When Humans Aren't Optimal: Robots that Collaborate with Risk-Aware Humans (2001.04377v1)

Abstract: In order to collaborate safely and efficiently, robots need to anticipate how their human partners will behave. Some of today's robots model humans as if they were also robots, and assume users are always optimal. Other robots account for human limitations, and relax this assumption so that the human is noisily rational. Both of these models make sense when the human receives deterministic rewards: i.e., gaining either $100 or $130 with certainty. But in real world scenarios, rewards are rarely deterministic. Instead, we must make choices subject to risk and uncertainty--and in these settings, humans exhibit a cognitive bias towards suboptimal behavior. For example, when deciding between gaining $100 with certainty or $130 only 80% of the time, people tend to make the risk-averse choice--even though it leads to a lower expected gain! In this paper, we adopt a well-known Risk-Aware human model from behavioral economics called Cumulative Prospect Theory and enable robots to leverage this model during human-robot interaction (HRI). In our user studies, we offer supporting evidence that the Risk-Aware model more accurately predicts suboptimal human behavior. We find that this increased modeling accuracy results in safer and more efficient human-robot collaboration. Overall, we extend existing rational human models so that collaborative robots can anticipate and plan around suboptimal human behavior during HRI.

Overview of "When Humans Aren't Optimal: Robots that Collaborate with Risk-Aware Humans"

The research presented in the paper "When Humans Aren't Optimal: Robots that Collaborate with Risk-Aware Humans" addresses the challenge of developing collaborative robots that can efficiently and safely interact with human partners in environments marked by uncertainty and risk. The focus is on extending existing human decision-making models to account for suboptimal behaviors influenced by these factors, employing Cumulative Prospect Theory (CPT) from behavioral economics as a central framework. This theory is known for capturing risk-aware human behavior, offering a more nuanced model than the conventional Noisy Rational approach.

Key Contributions and Findings

• Cumulative Prospect Theory in Robotics: The paper proposes integrating CPT into the modeling of human partners, allowing robots to predict suboptimal human behavior better. This model considers the cognitive biases that lead humans to perceive rewards and probabilities non-linearly, and as a result, make decisions that often deviate from purely rational choices. By adopting a Risk-Aware perspective, robots can more accurately model human actions, particularly under conditions of risk and uncertainty.
• Simulation and User Studies: The paper provides empirical support for the proposed Risk-Aware model across two scenarios: autonomous driving and collaborative cup stacking. In these experiments, Risk-Aware robots demonstrated improved prediction accuracy of human actions compared to a Noisy Rational model. The research findings from autonomous driving scenarios, for instance, showed a significant improvement in the alignment of robot actions with human driving behavior, especially in situations where human actions were suboptimal due to risk aversion or seeking.
• Planning and Collaboration Improvements: Through simulations and user studies, the authors developed planning algorithms based on Risk-Aware models. The implementation results suggest that robots leveraging these models achieved safer and more efficient collaboration with human partners. In a practical collaborative task of cup stacking, Risk-Aware robots displayed enhanced anticipation of human decisions, resulting in smoother interactions around shared tasks and environments.
• Statistical Results: The paper offers strong statistical evidence to support the effectiveness of the Risk-Aware model. Across various configurations and risk levels, the Risk-Aware strategy consistently outperformed traditional models in aligning robot actions with human expectations and behaviors. For example, the risk-related autonomous driving task demonstrated that Risk-Aware models could achieve a significantly lower Kullback-Leibler divergence in predicting human actions compared to the Noisy Rational model.

Implications and Future Directions

The results have noteworthy implications for advances in human-robot interaction (HRI). By incorporating risk-awareness into robot algorithms, it is possible to enhance the behavioral congruence and safety of robots operating in shared environments with humans. Particularly in domains like autonomous driving and industrial collaboration, understanding and forecasting human decision-making under uncertainty can lead to more intuitive and harmonious human-machine partnerships.

Looking ahead, the research opens pathways to refine human behavior models even further by integrating additional contextual and psychological factors and extending these concepts into more complex, interactive environments. Future work might involve enhancing the data efficiency of the models to improve performance with limited human interaction data, or exploring the integration of long-horizon planning in variable domains.

In summary, the paper makes significant strides in human-robot collaborative systems, offering insights into optimizing interactions through behavioral models attuned to human risk-related biases. This advancement paves the way for more adaptable and understanding robotic systems capable of integrating seamlessly into the complexities of human environments.