Decoding (un)known opponent's game play, a real-life badminton eye tracking study Aditi Mavalankar, Snigdha Dagar, Kavita Vemuri* International Institute of Information Technology, Hyderabad, India. Email: aditi.mavalankar@students.iiit.ac.in , snigdha.dagar@research.iiit.ac.in, kvemuri@iiit.ac.in Abstract before executing the serve. This period is defined as the time taken to access task relevant cues and strategize Is the underlying cognitive processes different when playing with an opponent whose game play is familiar to that of an appropriate motor actions (Vickers, 1996). unknown? This study filtered the advance cues extracted by expert and amateur players when paired with an opponent whose game Previous studies on badminton which looked at play is familiar to that of an unknown opponent. Our data collected differences in experts versus novice game play using spatial in a real-life naturalistic game play conditions suggests that at the occlusion concept (Abernethy & Russel, 1987a,b) show the beginning of the game and for the first serve only the opponent's former exhibited better anticipatory behavior while the later torso is crucial for cues and as the game progresses the information needed more information for decision making. In the same from the feet seems to be sufficient for the expert players in study a video recording of game play was shown to novice contrast to the data from the amateur player. Subsequently the preparatory or quiet-eye period for the serves at the beginning of and expert players who wore a mobile eye tracker and the the game play was higher than for later serves for all player sets. fixation duration at five distinct regions – shuttle-in-flight, The preparation time for known opponents by expert players was opponent's arm,racket,head,face,legs – revealed that both higher for the first serve than for the unknown opponents but by groups have similar early fixations regions though time on the fifth serve the duration was negligible. Analysis of complete the racket and arm was more for the expert while it was the rallies show that post-serve attention allocation to opponent's head area for the novice. Secondly they report that the order racket and shuttle-in-flight is paramount in the play for both sets of of the fixation on the cues was not dissimilar. A similar opponents. Taken together the results of this investigation suggests study on tennis (Goulet et al., 1989) aimed to understand that expert player's visual attention was distinguishable to that of visual search pattern reported that focus was on an amateur player and expert players quickly decode unknown opponent's competence fairly early in the game play and follow shoulder/trunk of the opponent in the preparation stage and consistent pattern of visual search. The results from the then shifts to the racket during the execution phase while preliminary experimental data suggest the possibility of novices depend on more cues by using pre-recorded game understanding how humans employ dynamic pattern recognition play as stimulus. Singer et al., (1996) also used simulated models in visual-search. tennis play and found differences between skilled and non- skilled player in visual-search , reaction time and decision Introduction accuracy with non-skilled players fixating longer on the Data collected from real-life naturalistic conditions opponent's head and less systematic in the tracking of the in- provides insight into anticipation, prediction and rapid re- flight ball. It was also shown that player's ability to adjustments processes applied by players in a sport like anticipate opponent's intentions from postural cues is an badminton. Analysis of data like eye gaze collected from advantage (Rowe & McKenna, 2001) and a skill that is players engaged in real-time naturalistic game play can acquired over time by players. provide an accurate reflection of player behavior. Inferences on the underlying cognitive and motor skills can be derived Using a more sensitive eye tracker, Abernethy (1990) to a certain extent from two main indices, visual search conducted experiment with video recorded squash game patterns and fixation duration in the preparatory, play projected on a huge screen on the wall of the squash anticipatory and execution phases of the game play. In this court. The players were positioned in the court and the data study we report scanpath analysis reflecting the visual showed that experts fixate on head/arm more than on the search in later two phases with emphasis on the quiet eye racket as compared to novice players, from which they period in the preparatory phase. We do this by comparing inferred that experts are capable of eliciting advance cues data from three players paired against opponents with whom just from posture. The fixation times were not different for they have played before and others whose game play was the two groups and visual search pattern variation was not unknown. A wearable eye tracker (Tobii Glasses I) was evident. A meta-analysis of three decades of work (Mann used to collect saccadic eye movement and the fixation et al. 2007) which compared the attentional allocation of duration, which is an estimate of attention allocation at experts and novices report that the former have fewer long particular regions of interest important for game strategy, duration fixations translating to possible higher information though it was shown that other factors like stress can also extraction ( Williams et al., 1999) quicker. The quiet-eye influence fixations (Abernethy, 1988, 1990). The quiet-eye time , as another gaze behavior index, of experts was found period in our case is the preparatory phase of the player just to be higher when compared to less-skilled players across 211 studies on a wide range of domains including in sports from the three players were collected, as the data was (Mann et al 2007). Other studies on badminton looked at collected from naturalistic real-time game play, utmost the dynamic patterns from the players mutual spatial effort was made to ensure that opponent and player's spatial displacement within the court as an important variable for position in the court was constant through out the forehand- speed scalar product estimates which showed increased serves with minimal variation in the velocity of release of stroke variability to disrupt stable patterns ( Chow et al., the shuttle. The participants were allowed to continue the 2014). rally till one of them dropped a shot. After all the sets were completed the player was asked to rate the opponent's play. The ability to gauge the expertise levels of the opponent on a scale of 1-10. The rating was taken purely on the basis and play optimally is a strategy applied by even reasonably of their game play on that particular day. The fixation good players in elite sports like badminton. As in any duration above a threshold of 70ms and the scanpath was competitive setting the player needs to quickly gauge and analyzed at three phases: a) the preparatory b) actual decipher the opponent tactics from overt visual cues like execution of the serve by the player and c) the complete facial expressions, body posture and spatial position on the rally. court and also covert memory models formed from previous encounters with the opponent. In the absence of prior information, the player needs to quickly build the same early on in the game to win. An expert player's skill is based on the ability to analyze opponent’s strength and weakness and evolve response strategy accordingly. Of interest to cognitive research and to sports personnel is the dynamic process applied by expert players who are able to quickly decode an unknown opponents game play, which is the focus of the present study. Towards this we collected data in a open-to- sky mud badminton court in near natural game play conditions from 3 highly rated badminton players pitched against six known and unknown opponents. Methods Participants Two players (P1, P2) rated 9/10 with at least 6-7 years of Figure 1: The badminton court where the data was collected experience and having participated in inter-college and the heatmap sample from one of the recordings. tournaments and a third player (P3) who was rated 7/10 but has not taken part in any serious competitive matches were Data analysis paired with 6 opponents (O1,O2,O3,O4,O5,O6) each. Of the 6 opponents, O1,O2,O3 were comparable in expertise to P1 Eye movement data was recorded at 30 frames per second. and P2, while the rest of the opponents were fairly good The video recording from the eye tracker was analyzed players. O1 and O2 had played in practice sessions with P1 frame by frame using the Tobii's studio. Heat maps were and P2 while the other opponents game play was unknown. generated for each serve from preparation time of a serve to P3 had played with O1 and O3 before. The participants when the rally is dropped by either one of the players. were all in the age group of 18-23 years and right-handed. These heat maps provide a relative measure of the duration The known and unknown opponents were mixed to take of gaze of the player in the different areas of interest in the care of any habituation that might occur in the players. All scene. From the coordinates the fixation duration at each players had given their consent before taking part in the gaze position in the scan path was estimated with main experiment. regions of interest being the opponent's – torso, feet and racket and the shuttle for 4 serves out of the 10. The Procedure selected serves were the first, second, fifth and the eighth The eye movements were recorded from head-mounted for all the sets. The scanpath of the player is represented in tracking device from Tobii (Glasses 1, the form of state diagrams, wherein each fixation duration at http://www.tobii.com/), the recording unit connected to the a position of interest is a state and change in eye movements glasses is the size of a smart-phone and hooked onto the is the transition between the states. participants track-pants and hence allows for natural play. All the experiments were carried out in the same familiar open-air court in naturalistic conditions and the frame-grab Results in Figure 1 shows the court with heatmap overlaid from a The scan path before and just after the serve is analyzed recording. For each pair of participants 10 random serves from two views a) quiet eye during the preparatory phase, 212 and b) the visual search pattern as the game progresses. the second serve switching between feet, upper body and Four serves (s1,s2,s5,s8) for each pair of players was shuttle. Interestingly P3 did not shift attention to the analyzed to look at differences in salient cues a player opponent's body after executing the serves whereas P1 and gathers in order to strategize an optimal serve and the P2 fixate on the opponent for 4 opponents after executing variation across the serves. Table 1 lists the average the serve. fixation period of the first and last two serves in the preparatory phase period grouped for known and unknown opponents. The preparatory period is the time just before the serve is executed by the player. For the known opponents the average preparatory time was higher (824ms) than for the unknown (670ms) for the first serves while the time for second set of serves was slightly higher for the unknown(435ms) than the knows (332ms). The preparatory time for the first serves were higher than for the later serves for both set of opponents. Table 1: The average fixation duration in milliseconds for preparatory or quiet-time period for known and unknown players. serve: s1,s2 Serve: s5,s8 time (ms) time(ms) Known P1 940 515 P2 755 297 P3 779 185 Average 824 332 Unknown P1 721 494 P2 464 387 P3 827 426 Average 670 435 Figure 2: state diagram representing the scan path from preparation to execute a serve and just after. Red: s1 (serve The scan path data from the preparation to execution of 1), Blue: s2, Green: s5 and Brown: s8. a) P1 with an known the serve gives insights to the visual cues gathered by the opponent -O1 and an unknown -O5. b) P2's serve with player to plan the serve, predict the return and anticipate the known -O1 and unknown -O5 and c) P3's with known O1 response. Figure 2 is the state diagram representation for the and unknown -O5. The nodes/states color code: opponent 4 serves, for one known (O1) and an unknown (O5) (before serve) – dark grey. Opponent's feet (before feet) – opponent for each of the players. The scanpaths of all light grey. Opponent's racket – blank/white. Shuttle – opponents were analyzed but not included in the figure due lavender. Opponent (after serve) – green. to size issues. The higher rated player, P1's first landing fixation is the opponent upper body for all the unknowns The detailed scan-path diagrams of two competitive and for 1 known player of the 6 opponents for the first serve rallies of the eighth serve of players P1 and P2 is shown in (red, Figure 2a), from second serve on the first fixation was Figure 3, with opponents O1 and O5. As can be inferred the opponent's feet consistent across all opponents. from the sample set of data, at the beginning the players Attention was also allocated to the opponent's upper body attention is on opponent or opponent's feet but shifts to after executing the serve and tracking the shuttle in-flight opponent's racket and shuttle during the actual rally for 4 out of the 6 opponents. The first fixation position for especially when paired with a known and higher rated P2 (figure 2b) was the opponent's upper body for 3 opponent (O1) a trend that was noticed from the analysis of opponents – 2 unknown- in the first serve while for 2 others other rallies with known players. For unknown opponents (both unknown) it was the opponent's feet and for one it was the attention away from shuttle or racket was dependent on below the 70ms threshold, so not considered. As was the the rally duration and the type of shot hence, no consistent case with P1, from the 2nd serve the first fixation point was pattern was discernible. For the amateur player (P3) the the opponent's feet. In the case of the amateur player, P3, fixations were random shifting to opponent and feet during the first landing position was also the opponent's upper body the rally frequently. in 4 (one known and rest unknown) but gets random from 213 expressions or from the position of the feet the player could deduce the upper body stance or a combination of both. Additionally, the better players (P1 & P2) tend to look at Figure 3: State diagram representation of the scanpath the opponent after the serve for some the serves, which data for 2 rallies each of a) P1 and b) P2 players with O1 could be either a function of the type of serve which and O5. The action or response that triggers the state change requires the player to reconfirm the opponents body cues to (gaze change) is indicated by the lines connecting the states. gauge motion pattern or the time for response was longer due to in-air flight time of the shuttle and attention shifts to the opponent. The second gaze to the opponent could also Discussion and Conclusions be to fine-tune the game play by recording the current The aim of the study was to investigate cognitive and spatial location and plan to position the return at a location motor skill differences when a badminton player is paired further away. For the amateur player (P3), after the serve the against known and unknown players. The visual search opponent upper body was not tracked for any opponent patterns suggest a deviation in the first relevant visual cue which means the player is missing important overt and gathered by the player in the first serve as against covert cues. The landing fixation for ensuing serves was subsequent serves, a trend that is noticed when playing with either the feet or upper body, and hence no pattern is either known or unknown opponents. The explanation could followed in contrast to higher rated players P1 and P2. This be the need for player to 'seize' the facial expressions for could mean that amateurs have not evolved an optimal cues on anxiety, nervousness or casualness to make an scanpath or the ability to elicit visual cues. In studies estimate or guess the expertise levels (unknown opponent) comparing expert versus novice (Mann et al., 2007; or intensity for the current game play (known). For example Abernethy & Russell, 1987b) no difference was found in an easy casual countenance might be perceived to indicate a visual search, but the paradigm applied was to analyze fairly good player and cues. And the posture of the torso can players response to a serve whereas in our study the interest possibly give clues about the planned response especially was visual cognition applied by player to predict opponent's for unknown players in the beginning of the game play. response to a serve. Further experiments with amateurs After the first serve and the subsequent rally from the need to be conducted to validate our preliminary finding. second serve on, the first landing gaze is the opponent's feet for nearly all pairs which could be due to either the player Though the scanpath pattern from known and unknown ability to retain a memory of the opponent's facial opponents was almost similar the preparatory duration or quiet eye period shows that for known and unknown 214 opponents the first two serves were higher across all the Abernethy, B., & Russell, D.G. (1987b). The relationship three players and by the fifth serve the quiet time was between expertise and visual search strategy in a racquet significantly lower. This could be because post the first sport. Human Movement Science, 6, 283-319. serve the opponent's upper body is not allocated attention. Additionally the time period for the first serves for known Chow, Y.J., Seifert, L., Hérault, R., Chai, J.Y.C.S, Lee, opponent was higher than for unknown and the difference is C.Y.M. (2014). A dynamical system perspective to lower as the game progresses that is, by the fifth serve. The understanding badminton singles game play. Human longer eye time for higher rated known players when Movement Science 33,70–84 playing with equally skilled opponents could possibly mean a structured planning processes applied by these players or Goulet, C., Bard, C., & Fleury, M. (1989). Expertise even anxiety at the beginning of the play (Williams & differences in preparing to return a tennis serve: A visual Elliott, 1999). The differences in information processing information processing approach. Journal of Sport & from visual cues of the expert players (P1 & P2) and the Exercise Psychology, 11,382–398. amateur (P3) player is similar to the findings reported in novice versus expert comparison studies ((Abernethy & Mann, D. T. Y., Williams, A. M., Ward, P., & Janelle, C. Russell, 1987b). M. (2007). Perceptual-cognitive expertise in sport: A meta-analysis. Journal of Sport & Exercise Psychology, In conclusion, the experiment conducted in real-life 29, 457–478. settings and with very few motor control instructions means we acquired natural actions but it also threw up data Rowe, R. M., & McKenna, F. P. (2001). Skilled analysis challenges and confident assertions were not anticipation in real-world tasks: Measurements of exactly possible. Nevertheless from the current set of data, attentional demands in the domain of tennis. Journal of we can infer that players decode unknown opponents game Experimental Psychology: Applied, 7, 60–67. play fairly early in the game formulate patterns from visual cues efficiently. An observation we noted was the Singer, R. N., Cauraugh, J. H., Chen, D., Steinberg, G. immersive play by the participants due to fewer restrictions M., & Frehlich, S. G. (1996). Visual search, anticipation, in motor actions or game play mechanisms. Future work and reactive comparisons between highly-skilled and could consider more participants of national or international beginning tennis players. Journal of Applied Sport standards to set the baseline for preparatory visual search Psychology, 8, 9–26. pattern and fixation times. Models from the coaches or professional players can be used by players to optimize Vickers, J.N. (1996). Visual control while aiming at a far acquisition of covert and overt information. target. Journal of Experimental Psychology: Human Perception and Performance, 22, 342-354. Acknowledgments Williams, A. M., & Elliott, D. (1999). Anxiety, expertise, Funding acknowledgment: Partially funded under the and visual search strategy in karate. Journal of Sport & serious games project of the National Programme on Exercise Psychology, 21, 362–375. Perception Engineering Phase II, Department of Information technology (DIT)/DIETY, Government of India. References Abernethy, B. (1988). Visual search in sport and ergonomics: Its relationship to selective attention and performance expertise. Human Performance, 4, 205-235. Abernethy, B. (1990). Expertise, visual search, and information pick-up in squash. Perception, 19, 63–77. Abernethy, B., & Russell, D.G. (1987a). Expert-novice differences in an applied selective attention task. Journal of Sport Psychology, 9, 326-345. 215