The 2009 IEEE/RSJ International Conference on Intelligent Robots and Systems October 11-15, 2009 St. Louis, USA
Robot Jenga: Autonomous and Strategic Block Extraction
Jiuguang Wang, Philip Rogers, Lonnie Parker, Douglas Brooks, and Mike Stilman
Abstract This paper describes our successful implementation of a robot that autonomously and strategically removes multiple blocks from an unstable Jenga tower. We present an integrated strategy for perception, planning and control that achieves repeatable performance in this challenging physical domain. In contrast to previous implementations, we rely only on low-cost, readily available system components and use strategic algorithms to resolve system uncertainty. We present a three-stage planner for block extraction which considers block selection, extraction order, and physics-based simulation that evaluates removability. Existing vision techniques are combined in a novel sequence for the identication and tracking of blocks within the tower. Discussion of our approach is presented following experimental results on a 5-DOF robot manipulator. Index Terms Jenga, Entertainment Robot, Block Extraction, Game Strategy, Movable Obstacles, Pose Estimation
I. I NTRODUCTION ENGA [1] is a popular game marketed by Hasbro, consisting of 54 wooden blocks constructed into a tower of three blocks per level (Fig. 1). Players take turns identifying, removing loose blocks, and placing them on top of the tower. As the game progresses, the tower becomes increasingly unstable. The objective of the game is to build the tallest tower possible (in a single-player setting) or avoid being the player that collapses the tower (in a multi-player setting). Tower instability, uncertainty in pressure distribution, and imprecise force control are all known to be difculties for human players, where the world record is 40 and 2/3 levels. A robot manipulator that plays Jenga faces an even greater challenge, where the robot must also recognize and localize Jenga pieces as well as track their unexpected motion. Due to these complex physical requirements, the design of a robot Jenga system has received signicant attention in the robotics community. South [2] created a spring-based physics simulation model to identify a playing strategy, but simulated a portion of the Jenga tower and not the full 18-level structure. Ziglar [3] approached the block extraction from a mechanics point of view and analyzed the translational and rotational moves for the likelihood of disturbing other blocks in the tower assuming a piecewise planar model; the results are used to determine moves that do not disturb the conguration tower, forming a play strategy. Barnum [4] constructed a simulator based on the commercial NVIDIA PhysX engine which is capable of simulating the full tower, and analyzed its stability with particular considerations to the center of gravity and the forces of friction. Kroger, et al [5]
The authors are with the Center for Robotics and Intelligent Machines (RIM) at the Georgia Institute of Technology, Atlanta, Georgia, 30332, USA. Emails: {j.w, progers7, lonnie, douglas.brooks}@gatech.edu, mstilman@cc.gatech.edu
Fig. 1. A Pioneer 3-AT robot and the associated 5-DOF servo manipulator, with a standard Hasbro Jenga tower. Image courtesy of Hasbro, Inc.
created a robotic Jenga platform to demonstrate multi-sensor integration using a Staubli RX 60 industrial manipulator, four PCs, a 6D force/torque sensor, a 6D acceleration sensor, a laser triangulation distance sensor, and two CCD cameras for visual feedback. The system was able to construct ten additional levels (29 blocks) on top of an 18-level tower, using no strategy beyond the random selection of blocks. The main contribution of the paper is the development and evaluation of an autonomous and strategic Jenga system, with particular consideration given to the planning of block extraction. In this work, we use a three-stage planner for block extraction which explicitly tracks the extraction history and simulates tower stability using a physics engine. Using active contour-based block pose estimation and optical owbased tower movement detection, we do not assume a structured environment such as the painted blocks used in the Kroger platform. We demonstrate our approach using low-cost, off-the-shelf components including a ve degrees of freedom servo manipulator (shown in Fig. 1), two CMOS cameras, and a single desktop for all sensing and control tasks. While the performance obtained was not comparable to the Kroger platform, the results are promising given the drastic reduction in cost and resources. The remainder of the paper is organized as follows. Section II formulates the Jenga game as a planning problem and proposes several strategies for use with a robotic manipulator. Section III describes the hardware and software architecture for the robotic Jenga system. Section IV describes the results obtained by applying our system with a simple robot on a generic Jenga tower. Finally, Section V provides concluding remarks and directions for future work.

978-1-4244-3804-4/09/$25.IEEE

Fig. 2.

A physics-based simulation showing that a particular removal leads to a tower collapse.
II. S TRATEGY A. Overview We present three-stage strategy, coupled with a visionbased error detection scheme, for autonomous extraction of blocks within a Jenga tower. First, we limit our block selection to side blocks to maximize the number of removable pieces. Second, we select a particular block from a list of potential choices generated in the rst stage and identify a single block from a decision table. Third, a physicsbased simulation is conducted for the nal block selection to identify the feasibility of removing that particular block. If the removal is deemed feasible, the robot proceeds to extract the block using vision-based error detection to monitor tower stability and abort if appropriate. B. Removal strategy On a tower with three blocks per level, there are two types of blocks available for removal, a single center block and two side blocks. There are advantages and disadvantages to both types: removing the center block ensures the stability of the tower, since there are still two side blocks on that level to support the pieces on top; however, there are no longer any available blocks for removal on that level. Removing the side blocks pushes the tower towards instability, but maximizes the number of removable blocks. While our algorithmic strategy has no constraints on which blocks to remove, our platform adds some restrictions. The ve-axis Pioneer arm cannot simultaneously control all six degrees of freedom for the end-effector. In the general case, translation of a block implies its rotation about some axis. For the middle block this necessarily implies a disturbance to the neighboring pieces. For the side blocks, we select motions that couple lateral translation with rotation about the vertical axis. The robot grips the small rectangular end of the Jenga block, and rotates the piece to the side. Aside from servo error, our approach causes no additional forces on neighboring blocks. C. Block Removal Planner Based on the removal strategy, we generate a list of candidate blocks for removal at every step. The Block
Removal Planner (BRP) selects a block in this list based on heuristic evaluation, and passes the information to the Removal Feasibility Planner for further analysis. The BRP tracks the history of removal attempts in a decision table which is used as the basis for new removal choices. The decision table couples each removable block with a tower movement threshold. The movement threshold is initialized to a pre-dened value and updated as the game progresses. The threshold keeps track of how much the tower moved when the robot last attempted to remove the particular block. By tracking earlier block motion, BRP can rank blocks according to threshold values. Hence, it identies the block that is least likely to affect tower stability. For every successful block removal, an additional validation step is used to consider previously failed extractions. As a particular block is removed, the forces due to weight and friction acting on its neighboring blocks change. If one of the neighboring blocks was previously marked unremovable, then this change justies another attempt. Consequently, with every successful removal, we iteratively mark the neighboring blocks for a new attempt. D. Removal Feasibility Planner Once the Block Removal Planner returns a potential choice, the Removal Feasibility Planner (RFP) conducts a physics-based simulation to determine if this removal choice could potentially lead to tower collapse. If simulation results in a collapse as shown in Fig. 2, then the BRP is asked to provide an alternate choice and the process is repeated. We chose the Newton Physics Engine, which has previously been shown to produce accurate approximations of static friction, outperforming all other engines compared in [6]. Depending on the size and structure of the tower, the RFP typically requires several seconds of computation on a standard desktop computer to complete a simulation. Once the RFP veries that the block can be removed in simulation, the robot proceeds with the manipulation as described in Section III. During a block removal, the motion of the tower is continuously monitored, and a stop signal is issued in the event that the tower movement exceeds the threshold. If a stop occurs, then the nal movement

Strategy System Block Removal Planner (BRP)
Vision System 3D Block Pose Estimation Tower Motion Estimation
Block Feasibility Planner (BFP)
Control System Inverse Kinematics
Manipulation Error Detection

Arm Control

Fig. 3.
The Jenga architecture with strategy, vision, and control subsystems.
value is recorded in the decision table, and the BRP is asked to provide an alternate choice. The tower movement is determined using a vision-based technique, also described in Section III. III. A RCHITECTURE A. Overview Our robot Jenga player consists of inexpensive, off-theshelf components. Compared to the Kroger platform, which uses a Staubli manipulator, we use a low-cost ve-DOF Pioneer servo arm. Instead of multiple PCs, we use a single desktop computer (2.0 GHz, 2 GB of RAM) to make decisions about motion, control the manipulator and process all sensor data. Our system only employs two CMOS cameras for visual feedback and no additional sensors. Despite our limited equipment, the system operates successfully with standard Jenga blocks and minimal control over lighting. In contrast to the structured environments described in previous work we do not assume painted/marked blocks and instead use monocular pose estimation to identify, localize, and track blocks with respect to the robot arm. The software architecture for our Jenga system consists of two subsystems, each with two components. First, the vision subsystem uses active contours for pose estimation and optical ow for motion estimation. Second, the control subsystem applies closed form inverse kinematics and controls arm motion. The structure of this architecture is given in Fig. 3 B. Block Pose Estimation The pose estimation subsystem uses a CMOS camera facing the front side of the Jenga tower from an angle. It is tasked with determining the 2D pose of all candidate blocks on this plane with the assumption that the distance from the robot to the tower is known. During setup, a calibrated camera image is transformed using planar homography so that one tower side appears from the manipulators reference frame. The pose estimation problem can then be separated into 3 steps: (1) locating candidate blocks, (2) tracking blocks across time, and (3) calculating pose. This is accomplished

using existing techniques applied in a novel sequence, overcoming variations in block color, texture, and other uncertain variables. 1) Locating candidate blocks: First, the image is processed using four lters with intermediate results shown in Fig. 4). The image is converted to grayscale, then subjected to Gaussian blur (to reduce woodgrain), Canny edge detector [7], and dilation using a square kernel. The resulting binary image is searched for connected regions of appropriate area using the topological method outlined in [8]. Any regions found are used to initialize the block tracker. 2) Tracking blocks: The tracker is based on an active contour model [9]. Because the Canny edge detector provides reasonable contour initializations, it is again used as the external energy measure of the active contours. This is implemented as a minimization of the negative Canny edge value along the blocks contour, and has proven effective in our experiments. We apply a heuristic method to infer the location of missing blocks from the even-odd level structure of the tower and the presence of other tracked blocks on a given level. Block tracking is suspended when a block is occluded by the manipulator if either of the following conditions are met: Area: Given block area ab and expected area ae , a block is removed if |ab ae | > a Cross-ratio: given four corners C1. C4 , the cross ratio is equal to CR(C1 , C2 , C3 , C4 ) = (C1 C3 )(C2 C4 ) (C1 C4 )(C2 C3 ) (1)
Any four points are invariant with respect to their cross ratio under a perspective transformation [10] so for known block corners b1. b4 and estimated block image corners i1. i4 , a block is removed if |CR(b1 , b2 , b3 , b4 ) CR(i1 , i2 , i3 , i4 )| > CR (2)
3) Calculating pose: Due to the simplication of blocks being attached to one side of the tower, the pose calculations are reduced to calculating block position relative to the tower plane. Because of the planar homography calculated during setup, an image of the tower from the manipulators

(a) Original

(b) Canny edge

(c) Dilation

(d) Contour Fig. 4.
(e) Snake Intermediate results from ltering, used to locate the candidate blocks.

(f) Pose

perspective is known. Using a constant ratio of pixels to real units (cm), the x and y components are computed as follows: cm (3) worldx = imagex pixels cm worldy = imagey (4) pixels The block tracking and pose estimation results are shown in Fig. 4. C. Tower Motion Estimation Vision is also used to estimate tower stability as it relates to the motion of the tower while the robot attempts to remove a block. If the movement exceeds a threshold, a stop command is issued, and the movement value is recorded. To accomplish this, we compute the optical ow eld of the tower as seen by an overhead CMOS camera, and compute a movement value based on the magnitude of the eld. We divide the motion estimation problem into several stages. First, we select easily identiable features that can be tracked across frames. Next, we run the Lucas-Kanade algorithm to compute the optical ow eld with respect to the movement of the blocks. Finally, we calculate a movement value based on the magnitude of the eld. 1) Feature Selection: In order to compute tower motion we must rst select robust features that can be relied on to compute optical ow. Initially, we chose Harris corners [11] as features to track, but encountered signicant errors in the motion estimation, as many of the corners are not readily identiable between frames. Instead, we settled on the feature selection criterion described in [12]. Based on the Shi and Tomasi denition, good features to track in motion estimation are features that minimize a dissimilarity function while using afne motion as the underlying image change model. This has been shown to be particularly compatible with the Lucas-Kanade algorithm discussed below. In experimentation, we found that while the Shi and Tomasi denition returned fewer features, the quality of the keypoints was superior to that of Harris. We were able to produce consistent matching results with relatively low computation.

2) Optical Flow: The Lucas-Kanade algorithm [13] is one of the most widely used algorithms for estimating optical ow. A template with an extracted sub-region (a small window) of the image is locally displaced in an attempt to minimize the squared error between two images. The LucasKanade algorithm assumes three basic conditions, Brightness constancy: a pixel does not change in appearance between frames. In grayscale the brightness should not change. Temporal persistence: the motion of a feature window changes slowly in time. Spatial coherence: neighboring points in a scene, belonging to the same surface, have similar motions, and project to nearby points on the image plane. With controlled lighting, small movement, and standard tower shape, our system satises all three assumptions. To avoid the aperture problem that results from large motions and without using a large window (which breaks the small and coherent motions assumption), we run the Lucas-Kanade algorithm in a pyramidal fashion. In this approach, we construct a Gaussian image pyramid of several layers, with increasing resolution from top to bottom. The motion estimation from each preceding level is taken as a starting point for estimating motion at the next level. This provides a low-cost assurance of accurate motion estimation. 3) Movement Magnitude: Once the optical ow eld is calculated via Lucas-Kanade, we simply sum the ow eld to obtain a value representing the overall movement in the frame. This value is then used by the planner described in the previous section for error detection. The threshold is determined experimentally. If needed, the ow eld can also provide the dominant direction of ow, but this was not used in our platform. D. Arm Control The motion of our servo-based robot arm was necessarily position controlled. We found a geometric closed form solution for the inverse kinematics of the Pioneer manipulator. Due to the limited range of motion required for removing individual blocks from the tower, we elected to keep the
gripper orientation facing towards the blocks and only varied its position when approaching blocks. The low-level arm control was implemented in the Player Project [14], a popular middleware for a variety of physical and simulated robots. The Player system follows a serverclient architecture, in which a sever running the Pioneer driver continuously listens for motor commands. The control algorithm was implemented as a client to the Player server which sends the appropriate joint commands. In this architecture, the individual components can be separated to different processors or computer, in the case that a single process cannot reliably control the arm in conjunction with other computationally intensive vision components. IV. E XPERIMENTS & R ESULTS In the experiments conducted, four adjustments were made to the game of Jenga to simplify the system: Due to the limited length of the manipulator, a ninelevel tower was used instead of the full eighteen-level structure. Due to the limited width of the gripper, the side blocks are pre-pulled forward approximately 0.5 cm from the tower; if a wider gripper is used, the extraction could take place from the longer side of the blocks, making this step unnecessary. After a block is removed, instead of placing it on top of the tower, the block is simply dropped to the side to simplify the manipulation task (i.e., the tower is not reconstructed). If the tower is disturbed during the block extraction process, the out of place blocks are not corrected (where in the ofcial game, this is required). Fig. 7 shows an example run of the the block extraction process, demonstrating various aspects of the decision process discussed in the previous sections. There are six frames in the gure taken from the accompanying video: in the rst frame, a tower is constructed with eight side blocks prepulled. In the second frame, the gripper is in place, ready to extract the block. In the third frame, block is successfully extracted. In the fourth frame, a new block is selected by the BRP, and the gripper is in place. In the fth frame, the previous extraction attempt exceeded movement threshold and a new block is selected by the BRP. Finally, in the sixth frame, another block is successfully extracted from the tower. Trials are set up to test the performance of the system. Out of 20 trials, we recorded the number of total attempts, successful extractions, and aborts due to tower movement; the statistics are shown in Fig. 5 and Fig. 6. Nine of the 20 trials successfully extracted three blocks, two trials failed to extract any blocks, and one trial extracted ve blocks. We also calculated the average success percentage, i.e., in a single trial the number of blocks extracted divided by the total number of attempts (which includes aborts). Overall, approximately half of the trials have an average success percentage of 90% or above. In the various components we used, such as history tracking, physics simulations, reactive stopping, etc, the vision-

7 Number of Trials 5

Number of Successful Extractions

Fig. 5.

Number of successful extractions for each of the 20 trials.
7 Number of Trials 0 0% 10% 20% 30% 40% 50% 60% 70% 80% 90%
Average Percentage of Success
Fig. 6. Average number of successful extractions per trial out of 20 trials.
based pose estimation and movement detection were the most relevant and critical to the success of block extraction. The accuracy of the vision results directly affects the planning and manipulation tasks and when incorrect readings are obtained, the extraction is more likely to fail. We typically experienced two types of pose estimation failures: failure to locate a candidate block and failure to calculate the true pose of the block. While the former simply implies that a block will not be considered in the planners, the latter lead to more serious consequences, as an incorrect pose almost always resulted in manipulation failures. While we believe the physics simulation is a valuable component in a successful Jenga strategy, we found that it did not play a signicant role in our experiments and often did not catch instability in the reduced-size tower. The simulation can reliably capture simple errors, e.g., removing the only support block on a level; however, for other stability simulations, the inaccuracy in vision and in the simulation engine itself often prevents it from returning valuable results.
(a) Full tower setup with eight side blocks pulled (b) A new block is selected by the Block Removal (c) The block is successfully extracted from the 0.5cm forward. Planner. tower.
(d) A new block is selected by the BRP. Fig. 7.
(e) The previous extraction attempt exceeded the (f) The block is successfully extracted from the movement threshold. A new block is selected. tower.
Example block extraction run showing the decision making process and error detection. [2] M. South, A Real-Time Physics Simulator for JengaTM , Masters thesis, University of Shefeld, South Yorkshire, England, May 2003. [Online]. Available: http://www.dcs.shefeld.ac.uk/intranet/teaching/ projects/archive/ug2003/pdf/u0ms4.pdf [3] J. Ziglar, Analysis of Mechanics in Jenga, April 2006. [Online]. Available: http://www.cs.cmu.edu/afs/cs/academic/class/ 16741-s06/www/projects06/ziglar jenga.pdf [4] P. Barnum, Algorithmic Analysis of the Stability of Stacked Objects, 2006. [Online]. Available: http://www.cs.cmu.edu/afs/cs/academic/ class/16741-s06/www/projects06/BarnumAlgorithmic2006.pdf [5] T. Kroger, B. Finkemeyer, S. Winkelbach, L.-O. Eble, S. Molkenstruck, and F. M. Wahl, A Manipulator Plays Jenga, Robotics & Automation Magazine, IEEE, vol. 15, no. 3, pp. 7984, September 2008. [6] A. Boeing and T. Br unl, Evaluation of real-time physics simulation a systems, in Proceedings of the 5th International Conference on Computer Graphics and Interactive Techniques in Australia and Southeast Asia. New York, NY, USA: ACM, 2007, pp. 281288. [7] J. Canny, A computational approach to edge detection, IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 679698, 1986. [8] S. Suzuki and K. Abe, Topological structural analysis of digitized binary images by border following, Computer Vision, Graphics, and Image Processing, vol. 30, no. 1, pp. 3246, 1985. [9] M. Kass, A. Witkin, and D. Terzopoulos, Snakes: Active contour models, International Journal of Computer Vision, vol. 1, no. 4, pp. 321331, 1988. [10] R. Hartley and A. Zisserman, Multiple View Geometry in Computer Vision. Cambridge university press, 2003. [11] C. Harris and M. Stephens, A combined corner and edge detector, in Alvey Vision Conference, vol. 15, 1988, p. 50. [12] J. Shi and C. Tomasi, Good features to track, in Computer Vision and Pattern Recognition, 1994. Proceedings CVPR94., 1994 IEEE Computer Society Conference on, 1994, pp. 593600. [13] S. Baker and I. Matthews, Lucas-Kanade 20 Years On: A Unifying Framework, International Journal of Computer Vision, vol. 56, no. 3, pp. 221255, 2004. [14] B. Gerkey, R. Vaughan, and A. Howard, The player/stage project: Tools for multi-robot and distributed sensor systems, in Proceedings of the 11th International Conference on Advanced Robotics, 2003, pp. 317323.

V. C ONCLUSIONS In this paper, we have presented the design and evaluation of an autonomous Jenga robot, using low-cost, off-the-shelf components. The main contributions of this work are the strategies in a Jenga game, which consist of block removal, tracking extraction history, and the physics behind tower stability. We have greatly reduced the number of sensors required for block extraction compared to previous implementations, and given the limited resources available, the results are highly encouraging. Some immediate steps towards a better Jenga system can be taken in the vision and physics domain, as was discussed in the previous section. A more advanced manipulator could be used to remove many of the limitations and assumptions. More sophisticated pose estimation and movement detection could greatly reduce the failure rate for the manipulation process. Finally, the integration of force feedback may further reduce failures due to tower movements and hence enhance playing capabilities. ACKNOWLEDGMENTS We are grateful to Dr. Ayanna Howard for providing the equipment used and Dr. Jim Rehg for his advice in the vision components of this project. We also thank Peter Barnum, Chung Hyuk Park, Hae Won Park, Sung Hyun Park, Sekou Remy, and Stephen Williams for their valuable comments and suggestions. This work was done as part of the fall 2008 course Robot Intelligence: Planning in Action at Georgia Tech. R EFERENCES
[1] (2008, December). [Online]. Available: http://www.hasbro.com/games/ family-games/jenga/

For the purpose of this Decision, 'electronic games' are console games, handheld electronic games and personal computer (PC) games. 'Traditional games' are defined as all games excluding electronic games. There are many clear differences between electronic games and traditional ones. Currently, there is little overlap between the suppliers of electronic games and the traditional toy manufacturers. Nintendo, Sony and Sega Enterprises Ltd dominate the market for electronic games. Electronic games are often supplied through different retailers, such as specialist electronic games retailers, audiovisual retailers or electrical goods retailers. Many such games require expensive hardware before they can be used. They are sold at price points that are much higher than those associated with most traditional games. Research conducted for Hasbro by Griffin Bacal states:
'principles of this [electronic games] differ from traditional board game play: mainly solo play manual dexterity/skill pace, speed mastery/control fast moving visual images Office of Fair Trading 16
visual and sound elements integrate as enhancers of excitement/ reward.'
The research goes on to say 'Handheld electronics deliver similar styles and types of game play, but are generally viewed as separate additional items not substitutes.' A comment from Hasbro in a Key Note report supports the view that traditional games form a separate market from electronic games. It believes board games will remain popular even in the Internet age, stating that 'the social interaction that board games bring is unique to home entertainment.'18 RELEVANT GEOGRAPHIC MARKET
As noted in paragraph 22 above in this case the OFT is under no obligation to come to a conclusion as to market definition for the purposes of establishing an infringement of Chapter I of the Act. However, market definition is the first step in assessing penalties. The OFT considers that it is unlikely that the market can be defined more narrowly than national. If a wider geographic definition were adopted this could have the effect of increasing the parties' relevant turnover and therefore penalty. For the purposes of calculating penalties the OFT is proceeding on the basis that the relevant market is that for toys and games in the UK. CONCLUSION
In the circumstances of the present case, the OFT does not consider it necessary to choose between the wider definition of all toys and games or the narrower definition given below of separate markets for each separate category. It is not necessary in this case to arrive at a precise definition in order to demonstrate an infringement of the Chapter I prohibition. However, the calculation of level of penalties depends partly on definition of the relevant market.The OFT has taken a narrow view of the market which results in penalties which are lower than if a broader definition had been adopted. Therefore, for the purposes of this Decision and in particular for the purpose of assessing the level of penalties the OFT has considered the relevant turnover of the parties in each of the following 10 categories of toys and games: 1) Infant and pre-school 2) Infant Pre-school

In this situation, Hasbro faces the risk that some of its products may be delisted in favour of those of other toy suppliers, own-brand offerings or even different products. 43 As is apparent from the evidence referred to below, in the period up to and including 1998 Hasbro management were aware that retailers were unhappy with the margins they were receiving and believed that this was mainly caused by [*].24 In response, they put together a number of initiatives, which were in operation from 1999 onwards. These initiatives were designed to improve retail margins overall and consisted mainly of what was known by Hasbro as a 'pricing initiative' and a 'listing initiative'.25 The listing initiative offered rebates in return for the listing of certain Hasbro products which were threatened to be delisted. The pricing initiative involved maintaining retail margins on Hasbro's toys and games range by persuading retailers to keep to recommended retail prices ('RRPs').26 In his witness statement, David Bottomley, a Sales Director at Hasbro, says that it was:
' a pricing initiative under which Hasbro would try to get retailers to list at RRPs.'27
Towards the end of 1998 and the beginning of 1999, discussions took place between Hasbro's sales team and buyers of Argos and Littlewoods over Hasbro's initiative and adherence to RRPs. The involvement of Mike McCulloch, Head of Marketing and Sales at Hasbro, David Bottomley and Mike Brighty, Hasbro Sales Directors, as active participants in setting up the initiative, as set
Report produced by Littlewoods plc 'Investigation into the Market for Toys: Points of General Clarification for the Office of Fair Trading' (21 December 2001). Witness statement of David Bottomley, paragraph 6; witness statement of Neil Wilson, paragraphs 4 and 5; witness statement of Ian Thomson, paragraph 14. Other initiatives included the introduction of clearance merchandise, an FOB programme and reducing cost prices (witness statement of Neil Wilson, paragraph 9). Witness statement of David Bottomley, paragraphs 8, 9 and 48; witness statement of Neil Wilson, paragraph 9; witness statement of Ian Thomson, paragraphs 33, 34 and 38 to 47.

'Argos (and other retailers) were asked by Hasbro whether they were happy to match Hasbro's RRPs. Argos said it was prepared to match RRPs as long as it was not undercut by its competitors. Andrew Needham [Argos's buyer of boys' toys] was certainly aware that Hasbro was communicating with retailers with a view to increasing margins by moving towards RRPs. I know this from conversations I had with him, including when he would pick up the telephone, say that he had seen an Action Man product for, say, 2 less than the RRP, and could Hasbro do anything about it. His purpose in calling me was that he wanted Hasbro to persuade the retailer to go back to RRP or, if we could not do that, to tell him so he could take account of that in his pricing. It was clear from this that he knew that Hasbro was persuading accounts to go to RRP. Sue Porritt [Argos's Senior Buyer/Trading Manager] felt it was great that Hasbro could help maintain retail price stability, but said that Argos would react if it was undercut in order to remain competitive. By 'retail price stability', I meant retailers going out at the same price, i.e. Hasbro's RRP. There was a two-way dialogue between Hasbro and Argos on pricing intentions. Whilst it is fair to say that Hasbro led this dialogue, it was not a question of Hasbro forcing this information from Argos or Argos not being clear about what this information was to be used for. There was an open dialogue, although Argos would only give an indication, not a guarantee, and would sometimes change a price without any consultation with Hasbro. Argos knew that Hasbro was talking to its other accounts in order to increase retail margins and was using indications received on prices for this purpose.'131
There is also other strong evidence that Hasbro agreed to fix prices with Argos. The paper prepared for the meeting between Hasbro and Argos on 17 February 1999 refers to 'Dialogue opened to stabilise RRP's' (see paragraph 51 above). An internal Hasbro report of a meeting between Neil Wilson and Sue Porritt on 29 March 1999 states: 'Hasbro's retail pricing strategy to increase trade brought [sic] in margin was discussed. Sue understands our strategy but categorically stated that Argos will react to competitor pricing ' (see paragraph 52 above).
Witness statement of Neil Wilson, paragraphs 10, 32, 44 and 69 (see further at paragraphs 22 to 25, 30, 46, 51 and 67 of his witness statement). This is supported by the witness statements of David Bottomley (at paragraphs 16, 19, 25, 28, 33, 46 and 48) and Ian Thomson (at paragraphs 65, 69, 86 and 98). Office of Fair Trading 54
Another internal Hasbro report of a meeting between Neil Wilson and Sue Porritt on 9 December 1999 states: 'SP was very positive about the new terms and the impact they will have on Argos business. It is crucial that we can maintain retail price stability across our key brands so that the plan can succeed ' (see paragraph 65 above). 112 In addition, Ian Thomson's internal e-mail of 18 May 2000 (see paragraph 67 above) states:

Office of Fair Trading 62
supported by the contemporaneous documents that are referred to in this Decision, there is no documentary evidence in support of those statements relied on by Littlewoods and Argos that contradict this view. The OFT takes the view that more weight should be placed on the statements, or parts of statements, supported by contemporaneous documents than on those not so supported.
Review of contrary evidence
In their representations, the parties have claimed that the OFT has not given sufficient consideration to the evidence that contradicts its case. However, in taking this Decision, the OFT has considered fully the entire body of evidence it has gathered, taken as a whole. As already indicated in paragraphs 15, 16 and 137, where there are clear contradictions in the oral statements, the OFT has relied on those statements supported by other evidence, in particular contemporaneous documentary evidence. Below, the OFT has specifically considered the oral statements taken from Hasbro and Littlewoods employees by the OFT (see paragraph 14) which might seem to contradict its case. The three Littlewoods employees interviewed by the OFT all denied that there was an agreement with Hasbro and Argos to fix prices. Lesley Paisley, Index buying manager, and Alan Burgess and Alan Cowley, both buyers of toys at Littlewoods, all stated that Littlewoods did not inform Hasbro of the prices it intended to charge and did not commit to any prices suggested by Hasbro. This is also the conclusion of a report that Littlewoods produced after an internal investigation. Hasbro's Head of Sales and Marketing, Mike McCulloch, also denied that there was an agreement with Argos to fix prices, nor did he concede that there was an agreement with Littlewoods. He stated that the e-mails which the OFT uses as evidence, 'look worse than they actually are' and show an 'over zealous approach by the account managers'. According to Mike McCulloch, Argos priced how it wanted and Ian Thomson, Hasbro's account manager for Littlewoods, could not possibly have guaranteed Argos's prices to Littlewoods. This seems also to be asserted by other Hasbro employees, who mentioned that they were not certain about the prices that Argos and Littlewoods would use until their catalogues were published. This has been construed as evidence of the non-existence of the agreement. However, in the OFT's view the Littlewoods employees and Mike McCulloch are contradicted by several other Hasbro employees, who were more closely involved in the Argos and Littlewoods accounts than Mike McCulloch. David Bottomley, a Hasbro Sales Director who was in the management hierarchy between Mike McCulloch and the account managers, states that '[w]hat existed between Hasbro and Argos and Hasbro and Littlewoods was an understanding

With regard to any actual improvements in margins, their existence suggests that they were driven by better buying terms from combining the sourcing of Argos's products with GUS's home shopping channel and by cost cutting
Office of Fair Trading 75
measures arising from the integration of the distribution arms of Argos and home shopping. They even suggest that the company intended to discount prices by 10 per cent in its home-shopping catalogue. 184 Hence, it is far from clear how such announcements could lead to a generally held belief that Argos would henceforth cease to price aggressively and start to follow RRPs. The consequent and largely simultaneous move to RRPs for certain Hasbro products, for example Action Man and core games, is not easily accounted for by reference to these cuttings. The only exception to this has been provided by Hasbro in its written representations. Hasbro quotes a commentator as stating in 'Housewares' of 1 September 1998: 'GUS hints that it may review Argos's price led marketing. During the takeover it derided Argos's recent 'the cheapest just got cheaper strategy' as folly.' This is the only press comment produced by Hasbro. Since none of the press comments submitted by Littlewoods refers to an intention by Argos to move away from its discounting strategy, the OFT does not find the existence of only one conflicting press comment persuasive. The GUS take-over of Argos occurred in April 1998 and Littlewoods states that the desire to improve margins dated from before that time to the period when Stuart Rose took over as Argos's Chief Executive. This would coincide with the preparation of the Argos Autumn/Winter 1998 catalogue. However, the move to RRPs is not generally seen until late 1999/early 2000. Hence, it is not clear how either of the earlier events could have sparked the change in policy that took place more than a year later. The OFT notes that the statements from the Littlewoods buyers of toys and also those buyers in other categories are ambiguous about a change in Argos's policy and Littlewoods's reaction and do not credibly demonstrate that any such change was the result of the GUS take-over. The Littlewoods buyers of toys indicate that they expected that Argos's announced intention after the GUS take-over to seek more margin would result in a move towards RRPs (except Andrea Gornall who does not mention any change in Argos's policy). However, the buyers contradict each other as to when Argos stopped its aggressive discount policy. John McMahon states that the policy change occurred in late 1998/early 1999, and Alan Burgess says the change was visible in the A/W 1999 catalogue, for which prices would have been established in early 1999. However, according to Lesley Paisley, Peter Edmonds and Katharine Runciman, Argos changed its policy only in late 1999 or 2000. Lesley Paisley, Littlewoods Buying Manager for toys, states that the policy change was visible no earlier than in the A/W 2000 catalogue (i.e. the

See also the witness statement of David Bottomley, paragraph 34. Office of Fair Trading 83
Katharine Runciman does not recall receiving the e-mail of 18 May, despite the fact that it purports to tell her about the pricing intentions of her closest competitor. Similarly, Phil Riley is unable to recall receiving the e-mail. His explanation for this is that it did not concern any of his products so he would probably have deleted it. However, as he was responsible for boxed games and the e-mail states quite clearly that 'Games and Action Man prices will continue to be adhered to', with 'Games' emboldened to make it stand out, the OFT finds his statement unconvincing. Alan Burgess too fails to recall the e-mail of 18 May, even though he observes in his statement that it is an unusual e-mail and admits that it looked as if either he or an assistant had ticked the e-mail presumably to check the prices against those Littlewoods were proposing to charge. The OFT also notes that he or an assistant was interested in these prices since Alan Burgess admits they were ticked. When so many of his colleagues state that they would not have trusted such information, this suggests that Alan Burgess, or perhaps an assistant, did attach some weight to them. Finally, while Alan Cowley does remember receiving the e-mail, he has no recollection of the 'conversations' referred to therein and says that he attached no importance to the contents. The OFT is surprised that Alan Cowley has no memory of such conversations, when he gave a detailed account of one such conversation earlier in his statement, i.e. discussions about the price of the Tweenies doll. Furthermore, whilst he states that he attached no importance to the contents of the e-mail, he does admit that, after the OFT began investigating, he checked the prices on the e-mail and that the products in his control in that e-mail did go out at the prices specified. The OFT is not convinced by the other argument raised by Littlewoods in respect of this evidence. The fact that the prices of four out of the 17 products (not including the many products included in the Action Man and games categories) were not actually applied by Argos (see further paragraph 78 above) does not constitute evidence that there were no agreements to adhere to RRPs or to fix prices. For example, it is clear with at least one of these products that subsequent conversations between the relevant buyers and their contacts at Argos and Littlewoods led to the agreed retail price for the product being changed: Interactive Pikachu was retailed by both companies at 23.75 in their Autumn/Winter 2000 catalogues. That it was Argos's intention to price this product at 23.75 was confirmed by Neil Wilson to Ian Thomson in a response to this e-mail on 25 May 2000 (see paragraph 75 above). The OFT notes that in its Autumn/Winter 2000 catalogue Littlewoods itself charged the prices as listed in the e-mail of 18 May 2000 for 15 out of the 17 products (see paragraph77 above). The remaining two products were Interactive Pikachu, which it priced at 23.75 in accordance with Neil Wilson's e-mail to Ian Thomson, and Gardens

Office of Fair Trading 84
Galore, which Thomson's e-mail had indicated would not be listed by Argos and which Littlewoods priced at 24.99 instead of 19.99. 221 Furthermore, the OFT is not required to demonstrate that both parties complied with the agreements in all its aspects throughout their duration to demonstrate that any such agreements exist, as the agreements had the restriction of competition as their object (see at paragraph 128 above). This is clear in the European jurisprudence, for example in a recent judgment of the European Court of First Instance where it stated that 'it is clear from case-law that, for the purposes of applying Article [81](1) of the Treaty, there is no need to take account of the concrete effects of an agreement when it is apparent that it has as its object the prevention, restriction or distortion of competition .' (see also at paragraph 256 below).167 INTERPRETATION OF THE E-MAIL OF 28 DECEMBER Representations: Littlewoods says, based on Alan Cowley's witness statement and that of Lesley Paisley, that the e-mail of 28 December (see paragraph 84 above) was a one-off that does not provide evidence of an agreement. Alan Cowley describes the circumstances surrounding the exchange, which appears to relate to a last minute reduction in the wholesale price and hence the RRP of the product. Alan Cowley was concerned that he would not be able to reduce the price in his catalogue and refers to the dire consequences of being undercut by Argos. OFT's response: The OFT accepts that none of the first part of the e-mail is particularly contentious. However, it is the last statement in this e-mail that the OFT is relying upon as part of its evidence of the existence of an agreement. The OFT finds it difficult to see what alternative meaning can be given to this final statement 'I will not elaborate on the consequences if we had been unable to do so, resulting in our being undercut by Argos and other High Street outlets, especially when you had earlier been so insistent that we all went out at the same price!' (emphasis added). Alan Cowley's witness statement does not seem to provide an alternative explanation for the statement; all it endeavours to do is to play it down and explain it away by saying that he used the term 'insistent' to exaggerate the situation. Lesley Paisley attempts to further downplay Cowley's words by saying that [*]. It should be noted that her comments on this issue
Cases T-202/98, T-204/98 and T-207/98 Tate & Lyle etc v Commission [2001] ECR II02035, paragraphs 72-73. See also European Court of Justice, Case C-277/87 Sandoz Prodottie Farmaceutici v Commission [1990] ECR I-45, paragraph 15. Compare the Belgian Roofing Felt Cartel case, where the European Court of Justice held that a pricing cartel existed despite the fact that the prices fixed for new products were not observed in practice (Case C-246/86 BELASCO v Commission [1989] ECR 2117, paragraph 15). Office of Fair Trading 85

' [It] must be borne in mind that in assessing an agreement under Article [81](1) of the Treaty, account should be taken of the actual conditions in which it functions, in particular the economic context in which the undertakings operate, the products or services covered by the agreement and the actual structure of the market concerned unless it is an agreement containing obvious restrictions of competition such as price-fixing, market sharing or the control of outlets .'172 (emphasis added) 'It is clear from case law that, for the purposes of applying Article [81](1) of the Treaty, there is no need to take account of the concrete effects of an agreement when it is apparent that it had as its object the prevention, restriction or distortion of competition within the common market (Case T-142/89 Boel v
See, for example, the European Commission Decision in Pre-Insulated Pipe Cartel ([1999] OJ L24/1, 30 January 1999). Cases T-374/94 etc European Night Services v Commission [1998] ECR II-3141, paragraph 136. Office of Fair Trading 92
Commission [1995] ECR II-867, paragraph 89; Case T-152/89 ILRO v Commission [1995] ECR II-1197, paragraph 32).'173 'The Court points out that, in order to find that an agreement is contrary to Article [81](1) of the Treaty, it is not necessary to establish that the agreement in question had an anti-competitive effect. A finding that an agreement pursued an anti-competitive object is sufficient for it to be declared contrary to Article [81](1) of the Treaty .'174 ' a concerted practice is caught by Article 81(1) EC, even in the absence of effects on the market.'175
In addition, the OFT cannot accept that price-fixing agreements involving, among others, the UK's biggest toy and games supplier, the biggest retailer and another major retailer of those products did not have as their object or effect an appreciable restriction of competition, even where applied to market sectors where Hasbro's share may not have been high. PENALTIES
The representations of Littlewoods on penalties are addressed in part V of this Decision.
Response to the representations made by Argos on the original rule 14 Notice

THE NATURE OF THE MARKET

Representations: Argos claims that the statement in the original rule 14 Notice that the market is reliant on branding and film tie-ins is not true for all categories of toys. Hasbro is very reliant on film success but has been poor at predicting demand for toys linked to films. [*] Toy sales are highly seasonal. Argos sees its main competitors as being [*]. Littlewoods has a much smaller share of the toy market but because they operate the same retail format consumers perceive Index and Argos to be direct competitors. It is easy for high street retailers to undercut the prices of catalogue retailers. OFT's response: The OFT does not accept all of the above representations. Even if it did, they do not go against the OFT's case that there were infringing agreements involving Argos. [*] the fact that Index and Argos use the same retail format differentiates them from other retailers and helps to explain why

Representations: When Argos entered the market its means of doing so was by seeking to offer the lowest prices on the High Street and other retailers came to view the Argos price as the one to match or beat. However, Argos is sometimes undercut and when it is it reacts. Following its acquisition by GUS in 1998 Argos [*]. Despite this it was still perceived to be the benchmark price by many retailers. OFT's response: These representations are accepted with the exception of that relating to the Argos [*] which is dealt with in detail below. It is an integral part
Office of Fair Trading 96
of the OFT's case that Hasbro's initiative needed the support of Argos if it was to be effective because Argos was seen as the price leader. 272 Representations: It is denied that Argos would make a commitment to follow RRPs only if Index did. No-one was aware that Hasbro was rolling out an initiative to persuade retailers to keep to RRPs nor were they aware that Hasbro had discussed any such initiative with any other retailers, including Littlewoods/Index. OFT's response: Argos relies on the statements of Maria Thompson, Argos's Commercial Director, and Andrew Needham and Vanessa Clarkson, both Argos toys buyers. These statements are contradicted by the documents and witness statements referred to in paragraphs 110 to 114 above. In addition, these statements are largely predicated on Argos's assertion that it had decided independently to move if it could towards [*]. This issue is dealt with in detail below. Representations: Catalogue retailers cannot modify their prices as easily as other high street retailers but prices do change although they cannot be increased. [*] OFT's response: The OFT accepts most of these representations. [*] DISCUSSIONS BETWEEN ARGOS AND HASBRO 276 Representations: Andrew Needham, the buyer in charge of Action Man and core games, does not recall any discussions regarding an initiative by Hasbro on those products. The 17 February 1999 meeting (see paragraph 51 above) between Hasbro and Argos was about [*]. The conversation with Maria Thompson when Mike McCulloch suggested he could ensure all retailers went out at RRP took place after that meeting. Key Argos personnel cannot recall statements by Hasbro to the effect that other retailers would follow Argos's price leadership. OFT's response: In so far as these representations are based on Argos employees stating that they are unable to remember things or were unaware of them, they can be given only very limited weight. The fact that the meeting on 17 February 1999 was intended by Argos to be about [*] would not have precluded discussion also of RRPs, and the e-mail from Sue Porritt of Argos indicates that a pricing strategy was in fact discussed (see paragraph 51 above). Given that Argos accepts that Mike McCulloch did make a suggestion about RRPs, it is likely that mention would have been made of Argos being price leader, which Argos accepts would have been a statement of the obvious. In addition, Andrew Needham's statement is contradicted by Neil Wilson, his counterpart at Hasbro. For example, Wilson says in his witness statement:

Office of Fair Trading 109
an infringement, other than the statements made voluntarily by the Hasbro employees. In addition, any denial of Hasbro is not sufficient to show that the Chapter I prohibition has not been infringed. The OFT has already responded to Hasbro's representations above (paragraphs 305 to 314 above), including any denials included in those representations. The OFT notes that Hasbro's application for leniency necessarily implies that Hasbro has admitted to a possible infringement of the Chapter I prohibition. In the OFT's view the most reasonable interpretation of the redacted version of the Hasbro representations that was given to Argos is that, while Hasbro denied infringing the Chapter I prohibition as set out by the OFT in the original rule 14 Notice, it did not deny that it had committed an infringement of some kind. ECONOMIC POWER OF HASBRO AND ARGOS 320 Representations: While Argos accepts that Hasbro did not have any ability to make Argos charge Hasbro's RRPs, Hasbro is wrong in asserting in its representations that the economic power was not with Hasbro, but with the retailers and Argos in particular. Hasbro possesses considerable bargaining/economic power in respect of the many 'must have' products that it supplies. In respect of other toy products, Argos accepts that it may have a high degree of bargaining/economic power. OFT's response: The OFT accepts that the economic power was not wholly with Argos, in particular in respect of Hasbro's 'must have' products, such as Action Man toys and Monopoly. However, Argos's position in the toy market (see paragraph 38) and the [*] share of Hasbro's sales accounted for by Argos (about [*] per cent) make it likely that even regarding these 'must have' products Argos has a significant degree of economic power. HASBRO'S POLICY REGARDING RRPs 322 Representations: In its representations, Littlewoods supports Argos's submission that there is no evidence that Hasbro has ever offered any incentives to Argos to price at RRPs. This is also supported by Hasbro's representations. There is a distinction between a supplier lawfully encouraging a retailer to keep to RRPs and unlawfully offering incentives to keep to RRPs. OFT's response: As stated in paragraph 267 above, it is not the OFT's case that Hasbro offered any direct incentives to Argos to induce it to keep to RRPs, although there clearly was an indirect incentive for Argos in the form of higher margins. This does not alter the OFT's conclusion on the basis of the evidence that Hasbro, Argos and Littlewoods entered into agreements to keep to RRPs (or, at times, another agreed price).

Section 39(1) of the Act provides for limited immunity from penalties for small agreements where the agreement is not a price-fixing agreement. The agreements between the parties in question are price-fixing agreements and therefore this limited immunity from penalties does not apply to the parties. In addition, the agreements do not fall within the category prescribed for the purpose of section 39(1) of the Act, as the combined turnover of the parties for the relevant business year exceeded 20 million (see part I.A).188 CALCULATION OF THE PENALTIES
In accordance with section 38(8) of the Act, the OFT must have regard to the guidance on penalties issued under section 38(1) of the Act when setting the amount of the penalty.189 Step 1 starting point
The starting point for determining the level of penalty is calculated by applying a percentage rate to the 'relevant turnover' of an undertaking, up to a maximum of 10 per cent. The 'relevant turnover' is the turnover of the undertaking in the relevant product market and relevant geographic market affected by the infringement in the last financial year.190 To be consistent with the Competition Act 1998 (Determination of Turnover for Penalties) Order 2000,191 the OFT considers that the last financial year is the business year preceding the date when the infringement ended. The OFT is of the view that the agreements between Hasbro, Argos and Littlewoods ended at the earliest on 15 May 2001, when the OFT visited Hasbro's premises in Uxbridge under section 27(3) of the
Regulation 3 of Competition Act 1998 (Small Agreements and Conduct of Minor Significance) Regulations 2000 (SI 2000/262).
'Director General of Fair Trading's Guidance as to the Appropriate Amount of a Penalty', March 2000 (OFT 423).

190 191

Paragraph 2.3 of OFT 423. Section 36(8) of the Act and SI 2000/309. Office of Fair Trading 122
Act (see further paragraph 388 below) and it is that date that the OFT will use in calculating the appropriate level of penalty in this case. 377 The actual percentage rate which is applied to the relevant turnover depends upon the nature of the infringement.192 The more serious the infringement, the higher the likely percentage rate. When making his assessment, the OFT will also consider a number of other factors, including the nature of the product, the structure of the market, the market share(s) of the undertaking(s) involved in the infringement, entry conditions and the effect on competitors and third parties.193 The damage caused to consumers whether directly or indirectly will also be an important consideration. An assessment of the appropriate starting point is carried out for each of the undertakings concerned, in order to take account of the real impact of the infringing activity of each undertaking on competition.194 The OFT considers price-fixing agreements to be among the most serious infringements caught under the Chapter I prohibition. 'The starting point for such activities and conduct will be calculated by applying a percentage likely to be at or near 10% of the 'relevant turnover' of the infringing undertakings.'195 The products concerned are consumer goods sold to a mass market through an established retail environment. They are very familiar, branded toys and games, that are aimed directly at children. Parents are under pressure to accede to the growing demands of children for the latest fad or trend. The heavy promotion and advertising of many such toys means that non-branded, cheaper alternatives are not viable substitutes for many parents. This also applies to 'old favourites', toys and games with long-established brand names such as Monopoly. Hasbro is a subsidiary of one of the two leading toy manufacturers in the world and supplies many of the leading brand names in toys and games, such as Action Man and Monopoly. These brands were among the first to be targeted in the price-fixing agreements and it is generally accepted that these are considered 'must have' products, with retailers believing that they cannot be seen as a viable toy retailer without stocking these brands, regardless of how low the margins are on such toys and games. This necessarily reflects the desirability of such brands to the consumer, with substitution to a non-branded alternative unlikely. While small-scale entry is clearly possible in the supply of toys and

Witness statement of Neil Wilson, paragraph 9. See also paragraph 126 of Ian Thomson's witness statement. Office of Fair Trading 130
The OFT has considered the evidence regarding who should be considered to have been an instigator or the instigator of the infringing agreements. As noted in paragraph 296 above, it is the OFT's view that discussions between Hasbro and Argos and Hasbro and Littlewoods took place over a period of time and that there evolved an understanding (which the OFT can accept was partly influenced by a desire on the part of both Argos and Littlewoods to increase profitability on toys and games by moving towards RRPs) that both Argos and Littlewoods would agree to adhere to RRPs on Action Man and core games on the understanding that the other would do likewise. In the circumstances the OFT accepts it would be difficult to point to a particular meeting or discussion as the occasion when the infringing price-fixing agreements came into being. However, on any reading of the evidence the OFT believes that it is sufficiently persuasive for it to find that Hasbro acted as an instigator of the infringements. Therefore the OFT has decided to increase the amount of the penalty by 10 per cent. Hasbro has made representations to the effect that it had in place a thorough and complete compliance programme and that its being ignored by senior management should not be regarded as an aggravating factor. In many cases the OFT is likely to find that the existence of an effective compliance programme is a mitigating factor and might make an appropriate downward adjustment to the level of penalty. In this case the existence of the compliance programme is offset by the fact that it was blatantly ignored at a very senior level within Hasbro and no adjustment is appropriate. However, the OFT is also aware of the remedial action taken by Hasbro's parent company, Hasbro Inc, following its discovery of the infringement. It has taken severe disciplinary action against the employees concerned and has stepped up compliance measures in its UK subsidiary by organising a specific competition law training programme for its senior management and sales staff and training in competition law for new staff. The OFT considers that in the light of these mitigating factors it is appropriate to reduce the amount of the penalty by 10 per cent. The OFT is normally minded to give a reduction in a penalty when a party has co-operated with its investigation. However, as Hasbro benefits from the leniency programme and as a condition of being granted leniency Hasbro agreed to co-operate fully with the OFT, the OFT does not consider that there should be an additional reduction in the penalties under this head to reflect general cooperation. The OFT believes that Hasbro committed the infringement intentionally. Hasbro's actions were intended to maintain the recommended resale prices of its products and Hasbro cannot have been unaware that this was likely to result in a

Argos's turnover in the categories affected by the agreement was as follows: boys' toys [*]; girls' toys [*]; plush [*]; games and puzzles [*]; creative [*]; hand-held electronic games [*]; infant and pre-school [*]. While the penalty has been calculated using the nonrounded turnover figures as provided by Argos, in this Decision only rounded figures are mentioned. Office of Fair Trading 133
Step 3 adjustment for other factors 417 The infringement enabled Argos to charge the recommended retail price for the Hasbro products concerned, with minimal risk of being undercut by its competitors. This allowed Argos to make higher margins on the Hasbro products concerned than it would have made without the infringement and thus to make considerable gain. However, arithmetical calculation of a gain should not form the sole or even the main means of marking the seriousness of an infringement except in the clearest cases (see paragraph 398). The OFT is satisfied that a penalty figure of [*] at this stage of the calculation is sufficient to act as an effective deterrent to Argos and others, in particular undertakings that might be considering engaging in price-fixing, and taking the factors of gain and deterrence together has decided not to increase the amount of the penalty at this step. Step 4 adjustment for further aggravating and mitigating factors 419 The OFT finds that Hasbro was an instigator of the infringing agreements. While there is some evidence that Argos was an instigator, there is no clear evidence against Argos in this respect and therefore it is not appropriate to make an adjustment to the penalty for Argos in respect of this aggravating factor. In recognition of Argos's full co-operation with the investigation the OFT has reduced the amount of the penalty by 10 per cent. As a result, there are no increases of the penalty for aggravating factors and the total percentage deducted from the penalty for mitigating circumstances is 10 per cent. The penalty for Argos is therefore determined at 17.28 million. Step 5 adjustment to prevent the maximum penalty being exceeded and to avoid double jeopardy 422 For the purposes of section 36(8) of the Act, 10 per cent of Argos's turnover (see paragraph 2) exceeds the level of the penalty by a very wide margin and the OFT does not consider it necessary to do the detailed calculations carried out for Hasbro at step 5. There are therefore no further adjustments to the penalty.
PENALTY FOR LITTLEWOODS Step 1 starting point 423 Littlewoods's turnover208 in the relevant product and geographic markets (i.e. UK markets for boys' toys, games and puzzles, girls' toys, infant and pre-school, plush, creative and hand-held electronic) in the financial year preceding the termination of the agreements (the financial year ended 30 April 2001) was [*].209 The OFT has set out its views generally about the seriousness of this infringement at paragraphs 376 to 387. With specific regard to Littlewoods, the OFT takes into account the very serious nature of the infringement (price-fixing) and its comments in those paragraphs regarding the nature of the products, entry conditions, damage to consumers and the effects on competitors. Although the position of Littlewoods in the retail toy sector is less important than the position of Argos, Littlewoods's share of the retail supply of traditional toys and games is significant. Littlewoods is a substantial and well known retailer in its own right. Despite Littlewoods's lower market share in the retail toy sector compared with Argos, Littlewoods is seen as Argos's main competitor in the high street catalogue sector. This is caused by the similarity of their outlet channel, the ease with which consumers can compare their prices because these are included in their catalogues, and their price-match guarantees. This means that Argos would not have taken part in the infringing agreements without the participation of Littlewoods. In the OFT's view Littlewoods would have been well aware that its participation in the infringing agreements was essential in order to bring Argos and its much larger market share within the scope of the infringement. It would also have known that other retailers would have been likely to follow Argos's prices since Argos is the acknowledged price leader in the market. Littlewoods's lower market share is not, therefore, a factor that should lead the OFT to find that its participation in the agreements should be viewed less seriously than that of Argos. Market share in any event is only one of the factors taken into account in assessing the seriousness of an infringement at step 1 and the OFT is in no