Abstract
Many everyday tasks require the ability of two or more individuals to coordinate their actions with others to increase efficiency. Such an increase in efficiency can often be observed even after only very few trials. Previous work suggests that such behavioral adaptation can be explained within a probabilistic framework that integrates sensory input and prior experience. Even though higher cognitive abilities such as intention recognition have been described as probabilistic estimation depending on an internal model of the other agent, it is not clear whether much simpler daily interaction is consistent with a probabilistic framework. Here, we investigate whether the mechanisms underlying efficient coordination during manual interactions can be understood as probabilistic optimization. For this purpose we studied in several experiments a simple manual handover task concentrating on the action of the receiver. We found that the duration until the receiver reacts to the handover decreases over trials, but strongly depends on the position of the handover. We then replaced the human deliverer by different types of robots to further investigate the influence of the delivering movement on the reaction of the receiver. Durations were found to depend on movement kinematics and the robot's joint configuration. Modeling the task was based on the assumption that the receiver's decision to act is based on the accumulated evidence for a specific handover position. The evidence for this handover position is collected from observing the hand movement of the deliverer over time and, if appropriate, by integrating this sensory likelihood with prior expectation that is updated over trials. The close match of model simulations and experimental results shows that the efficiency of handover coordination can be explained by an adaptive probabilistic fusion of a-priori expectation and online estimation.
Introduction
Humans coordinate manual interactions, like object handovers, with remarkable fluency. Whether passing a salt shaker or a heavy tool, we seem to intuitively know when and where to reach. This study investigates the underlying computational mechanisms of this coordination, specifically testing whether humans use a Bayesian "Causal Inference" framework to fuse prior expectations with real-time visual evidence.
For the field of robotics, understanding these mechanisms is crucial. If robots are to work alongside humans effectively, they must move in ways that are predictable and "fluent" to their human partners.
Experimental Methodology
The researchers conducted five experiments focusing on the receiver's reaction time (RT) during a cube handover task. The study utilized primary data collected via high-precision motion tracking: Polhemus Liberty (240 Hz) for human-human experiments and Intersense IS-600 (20 Hz) for human-robot experiments. Participants included groups of 12 to 20 right-handed university students.
The experiments compared human-human interactions against human-robot interactions, using both humanoid (JAST) and industrial (JAHIR) robots. A key variable was the robot's motion profile: "biological" (minimum-jerk) vs. "robotic" (trapezoidal) velocity profiles.
Key Insights and Results
The study's findings highlight the importance of biological motion and rapid cognitive adaptation:
- Kinematics Over Appearance: The robot's motion profile was more influential than its physical appearance. Humans reacted significantly faster to a robot moving with a "minimum-jerk" profile—which mimics human biological motion—than to one with a standard industrial trapezoidal profile.
- Rapid Adaptation: Humans are incredibly fast learners. In experiments with shifted handover positions, reaction times dropped significantly after just 2-3 trials as participants updated their internal "prior" expectations.
- Evidence Accumulation: Coordination is not a simple reaction to the start of a movement. Instead, receivers accumulate visual evidence over the course of the deliverer's trajectory. As certainty about the endpoint increases, the receiver initiates their reach.
- The Causal Inference "Switch": When a partner moves to an unexpected location, the brain experiences a conflict. The proposed Bayesian model suggests we "reject" our prior expectation and switch to pure visual tracking, explaining the initial spike in reaction time during unexpected events.
Discussion and Future Work
The research demonstrates that the efficiency of manual coordination can be explained by an adaptive probabilistic fusion of prior expectations and online estimation. Interestingly, participants did not show the same rapid adaptation to robots as they did to humans, possibly due to a lack of experience with robotic systems or the "unnatural" feel of the robot's joint configurations.
Future work aims to extend this framework to a "shared task representation" where the model accounts for the actions and expectations of both participants simultaneously, rather than just the receiver. This could lead to more intuitive and fluent human-robot collaboration.
Summary updated