Johannes Keck, Kazuyuki Sato and Rouwen Cañal-Bruland of Friedrich-Schiller-University Jena set out to close a persistent gap in the science of action understanding. Most previous experiments examined movement cues and contextual inference in isolation, often asking passive observers to make explicit judgments about pre-recorded actions. Real social life is nothing like that. Interacting partners respond to each other continuously, in real time, while the information available to them shifts from moment to moment. To capture this dynamism, the team turned to an unlikely but elegant experimental model: the children’s game rock-paper-scissors. As a zero-sum game with a cyclic, non-transitive payoff structure—rock beats scissors, scissors beats paper, paper beats rock—it combines genuine competition, multiple action alternatives and a rich history of play, all of which unfold within roughly a second and a half per game.

Ninety participants, paired into forty-five dyads, played hundreds of iterative games of rock-paper-scissors while wearing full-body motion capture suits studded with fifty-seven reflective markers. A sixteen-camera near-infrared system recorded their movements at 240 hertz, allowing the researchers to reconstruct the characteristic up-and-down wrist swings in exquisite detail. Each game was segmented into three phases corresponding to the first, second and final swing of the wrist, and eleven movement features—ranging from peak velocity and acceleration to trajectory skewness, positional variability and a measure of velocity unpredictability called velocity entropy—were extracted for every phase. After preprocessing, thirty-eight dyads with complete, reliably reconstructed trajectories entered the final analysis.

The first question was whether intention information is actually embedded in these naturally generated movements. To find out, the researchers trained a multinomial logistic classifier to map the phase-specific kinematic features onto the action each player ultimately performed. The classifier decoded the intended gesture with an accuracy of 47.6 percent, significantly above the 33.3 percent chance level. Crucially, this encoding accuracy did not differ between winning and losing trials, nor across the three movement phases. In other words, every player’s wrist leaked information about their upcoming choice, whether that choice would ultimately triumph or fail. The information was there for the taking, in every game, at every stage of the unfolding movement.

The surprise came when the researchers asked whether opponents actually exploited this information. Because explicit perceptual reports would have disrupted continuous play, the team derived an indirect, choice-based measure they call likely kinematic readout: if a player chose rock, for example, the analysis inferred that the player expected the opponent to throw scissors, and then tested whether the opponent’s kinematics on that trial actually supported that inference. Overall, likely readout exceeded chance, reaching 39.2 percent. But the outcome-dependent pattern was striking. On losing trials, readout was significantly above chance in every phase and rose steadily as the game progressed, peaking at 38.6 percent in the final swing. On winning trials, it never exceeded chance at all and actually declined to 31.1 percent by the end. Players who consistently followed the movement-derived signal across all three phases lost significantly more often than those who did not.

At first glance this seems paradoxical: why would reading the opponent’s body better make you lose? The authors argue that in fast-paced competitive settings with multiple possible outcomes, kinematic cues are noisier and more ambiguous than in the controlled, cooperative tasks used in most prior studies. Opponents may also deliberately disguise their intentions, converting kinematic information into kinematic disinformation, as research on deceptive movements in sports has repeatedly shown. Under such conditions, generating several competing kinematic predictions can reduce reliability despite high perceptual sensitivity, and observers may do better to complement—or override—movement evidence with contextual cues drawn from the game’s structure and history.

To test this idea formally, the researchers modeled each player’s contextual inference using four competing computational models drawn from two conceptual families. Action-based models predicted choices from sequential regularities in previous actions: a first-order Markov model tracking transition probabilities between consecutive gestures, and the cognitive hierarchy assessment model, or CHASE, which formalizes recursive strategic reasoning across levels of sophistication. Outcome-based models adapted predictions according to wins and losses: a win-stay/lose-shift model and a reinforcement-learning model implementing a Rescorla-Wagner update rule with softmax action selection. These model-derived posteriors were then fed, together with the kinematic measures, into hierarchical mixed-effects logistic regressions predicting single-trial winning.

The results were unambiguous. Reinforcement learning was the strongest positive predictor of winning: trials in which the model assigned high posterior probability to the player’s chosen action were significantly more likely to end in victory, with an odds ratio of 1.37. Win-stay/lose-shift contributed a smaller positive effect, whereas reliance on first-order Markov sequential regularities actually predicted reduced winning, suggesting that players who exploit simple action patterns become more exploitable themselves. On the kinematic side, richer intention information encoded in one’s own movement was modestly beneficial, but stronger behavioral alignment with the likely kinematic readout significantly reduced the odds of winning. Both information sources contributed unique predictive value, but context explained substantially more trial-to-trial variation in success than movement did.

The most compelling evidence came from conflict trials—decisive games in which the contextual model and the likely kinematic readout recommended different actions. Here, success can be attributed to one source or the other, and context won decisively. Following the reinforcement-learning recommendation more than doubled the odds of winning relative to following the kinematic signal, with an odds ratio of 2.33. Descriptively, participants won 60.6 percent of disagreement trials when they followed the reinforcement-learning recommendation, compared with only 42.7 percent when they followed the kinematic readout—a 17.9 percentage point advantage that significantly exceeded the benefits of the Markov, win-stay/lose-shift and CHASE models. An interaction analysis reinforced the picture: the behavioral influence of kinematic readout became increasingly detrimental as reinforcement-learning evidence accumulated, while the contribution of kinematic encoding itself remained additive across contextual states.

Taken together, the findings suggest that successful action understanding in dynamic social interaction is not a matter of maximizing sensitivity to bodily signals but of adaptively arbitrating between information sources. Movement kinematics consistently carried intention-related information, yet its behavioral value depended on the contextual state in which it was interpreted. The study also refines earlier neural work suggesting that tracking recent outcomes can impair performance in rock-paper-scissors: rather than contextual information being generally maladaptive, different forms of context have markedly different consequences, with reward-based learning outperforming simple sequential dependencies. The authors caution that their readout measure is an indirect behavioral proxy, and they propose that immersive virtual reality with systematically manipulated avatar kinematics could validate it more directly. For now, the message is clear: in the heat of real-time competition, the body talks, but the smartest listeners weigh what it says against everything else they know.

Subject of Research: Kinematic and contextual information integration in real-time competitive social interaction

Article Title: Context shapes action understanding in social interactions

Article References: Keck, J., Sato, K., & Cañal-Bruland, R. (2026). Context shapes action understanding in social interactions. iScience, 29(11), Article 117796. https://doi.org/10.1016/j.isci.2026.117796

Image Credits: AI Generated

DOI: Not provided

Keywords: action understanding, kinematics, social interaction, rock-paper-scissors, reinforcement learning, motion capture, action prediction, deception, computational modeling, decision making, contextual inference, iScience

The challenge is formidable. Multiple pursuit spacecraft must cooperatively capture an evasive target while operating under impulsive maneuver constraints and strict fuel budgets, producing an optimal control problem marked by strong nonlinearity and complicated constraints. The researchers note that the tools traditionally brought to bear on such problems each fall short in some way. Classical differential game methods suffer from high computational complexity and poor real-time performance, making them difficult to deploy on hardware with limited processing power. Numerical optimization approaches, meanwhile, are notoriously sensitive to initial guesses and struggle to handle constraints that persist over long durations. Deep reinforcement learning has shown genuine promise in some orbital scenarios, but existing algorithms lack the ability to proactively predict the actions of an opponent confronting them with an unknown evasion strategy, which limits how well the pursuit can be planned. The key question, therefore, became how to predict the evading spacecraft’s actions in multi-to-one scenarios and then optimize the cooperative strategies of the pursuers accordingly.

The team’s answer is a pursuit strategy construction method that fuses an evader action prediction network with the Deep Deterministic Policy Gradient algorithm, a well-established deep reinforcement learning technique for continuous control. The mathematical foundation of the work rests on the Clohessy-Wiltshire equations, the standard linearized description of relative spacecraft motion, which the researchers use to build a model of the multi-to-one impulsive orbital pursuit-evasion game in the Local Vertical-Local Horizontal reference frame. In this modeled world, both sides execute impulsive velocity-increment maneuvers at fixed time intervals, and each side faces constraints on both the size of each individual maneuver and the cumulative total velocity increment it may expend. Multiple pursuers cooperatively close in on the evader through finite sequences of these impulses, and the mission is declared a success the moment the distance between any pursuer and the evader falls below a preset capture threshold. Communication among the pursuers is itself constrained: information can be shared only within a certain proximity, adding a further layer of realism to the problem.

To train the pursuit policy, the researchers adopted a centralized training with decentralized execution framework, an architecture that has become popular in multi-agent reinforcement learning. During training, a single policy network is updated using global state information, but at execution time each pursuer relies only on its own local observations to generate impulsive maneuver commands. Crucially, all pursuers share the same network parameters, meaning that a fleet of chasing spacecraft can behave in a coordinated manner without needing a central commander in the loop during the actual engagement. The observation available to each pursuer is rich: it comprises the relative state with respect to the evader, the states of the two nearest fellow pursuers within communication range, and the predicted future state of the evader at the next time step, produced by the prediction network, for a total of 29 dimensions. The policy network maps this observation to a three-dimensional continuous impulsive velocity-increment command.

The reward structure that shapes the learning process is deliberately simple but effective. It combines a continuous reward for approaching the evader, a large constant reward for a successful capture, and an additional penalty for fuel consumption, so that the pursuing agents learn to be aggressive yet economical. Alongside the policy network, the team trained a separate evader action prediction network through supervised learning. Its input consists of the relative states of the three pursuers closest to the evader, and its output is a prediction of the evader’s velocity increment, effectively giving each pursuer a glimpse of the opponent’s likely next move. Importantly, this prediction network is trained using the global state information and the evader’s true action data collected during the pursuit policy training itself, so no additional data acquisition is required. All of the networks involved, including the policy network, the Q-value network, and the prediction network, are four-layer fully connected architectures in which the first three layers use the ReLU activation function and the output layer employs tanh.

The simulation campaign that validated the approach produced striking results. Over 200,000 training episodes, the reward trends of the pursuer and evader policies ran in generally opposite directions and stabilized after roughly 125,000 episodes, a signature that both sides had reached an approximate game equilibrium. When the researchers compared training curves for the pursuit policy with and without evader action prediction, the incorporation of prediction significantly improved both the convergence speed and the reward values achieved, while the loss curve of the prediction network itself converged after about 1,000 epochs. In test engagements, trajectories produced under three different prediction approaches, the learned prediction network, conventional Kalman filtering, and simply assuming the evader repeats its previous action, all achieved successful capture, but their accuracy differed markedly.

The prediction network proved decisively superior in tracking the evader’s future position. Its prediction error remained essentially below 0.3 kilometers throughout the pursuit-evasion process, with an average of just 0.137 kilometers, and it gradually decreased in the later stage of the engagement as the pursuers closed in. By contrast, the errors of the comparative methods approached 0.6 kilometers in the later stages, more than four times worse. The statistical picture was equally compelling. With three pursuers deployed, the capture success rate of the prediction-augmented method reached 0.95, an improvement of approximately 0.3 over the conventional DDPG approach. The researchers highlight that this advantage effectively compensates for the disadvantage of insufficient pursuer numbers, a point of real practical significance in a domain where every additional spacecraft carries substantial cost.

The study is also candid about the limits of the method. When the evader adopted a non-maneuvering strategy, the capture success rate climbed to 0.99, but performance degraded when the pursuers confronted unseen random or periodic evasion strategies, exposing a generalization bottleneck that is common to learning-based policies. The geometry of the opening position mattered as well: the initial distribution of the pursuers significantly affected the success rate, with an optimal initial distance of approximately 38 kilometers and an optimal enclosure area of approximately 2,300 square kilometers. The authors acknowledge that the approach currently relies on prior action data of the evader, and they outline future work on prediction methods that do not require such data, as well as extension to the even harder multi-to-many pursuit-evasion game problem.

For the wider space community, the significance of this work lies in its demonstration that foresight and learning can be combined to produce autonomous, fuel-aware, and cooperative pursuit behavior under realistic constraints. As activities in orbit multiply, from Mega-constellations to debris removal and on-orbit servicing, the ability of spacecraft to make fast, intelligent decisions in adversarial or uncertain encounters is poised to become a core capability. By showing that a supervised prediction of an opponent’s maneuver can be woven directly into a reinforcement learning pursuit policy, and by quantifying exactly how much that foresight is worth in terms of convergence speed and capture success, the Harbin Institute of Technology team has provided both a working method and a meaningful engineering reference for the autonomous decision-making systems that future space missions will demand.

The reliance on the Clohessy-Wiltshire equations places this work within a long tradition of orbital relative-motion analysis. These linearized equations assume a circular reference orbit and short engagement distances, which keeps the dynamics tractable enough for learning-based methods while remaining a reasonable approximation for proximity operations. The choice of impulsive maneuvers, rather than continuous thrust, mirrors the pulsed thruster firings common on real spacecraft, where each burn is short compared with the orbital period and fuel is measured in velocity increment.

The decentralized execution aspect of the training framework carries practical weight for on-orbit deployment. Because each pursuer generates its maneuver commands from local observations alone, the approach tolerates the intermittent communication links and latency that plague multi-spacecraft operations, where a central coordinator could become a single point of failure. The communication-range constraint built into the model reflects this operational reality.

The comparison against Kalman filtering is also instructive. Kalman filters estimate a target’s state from noisy measurements but do not anticipate deliberate maneuvering, so their predictions degrade precisely when an evader fires its thrusters. A learned predictor trained on maneuver data can instead capture behavioral patterns that a purely dynamical filter cannot represent. The reported average prediction error of 0.137 kilometers, against roughly 0.6 kilometers for the alternatives late in the engagement, quantifies this advantage in terms familiar to guidance engineers.

The acknowledged dependence on prior evader action data remains the principal caveat, since real adversaries may behave in ways no training set anticipated.

Subject of Research: A predictive learning-based pursuit strategy for multiple-to-one orbital pursuit-evasion games among spacecraft.

Article Title: A predictive learning-based pursuit strategy for the multiple-to-one orbital pursuit-evasion game

Article References: A predictive learning-based pursuit strategy for the multiple-to-one orbital pursuit-evasion game. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: orbital pursuit-evasion, spacecraft, deep reinforcement learning, DDPG, action prediction, Clohessy-Wiltshire equations, impulsive maneuvers, CTDE, autonomous decision-making, space security, Harbin Institute of Technology, Space Science and Technology