Online Bayesian Goal Inference for Boundedly-Rational Planning Agents
This addresses the problem of enabling machines to infer goals like humans do, which is incremental as it builds on existing inverse planning methods but with a focus on bounded rationality and online inference.
The paper tackled the problem of inferring an agent's goals from both optimal and non-optimal action sequences, presenting an architecture that models agents as boundedly-rational planners and developing the Sequential Inverse Plan Search algorithm, which outperformed Bayesian inverse reinforcement learning baselines in accurately inferring goals across domains with compositional structure and sparse rewards.
People routinely infer the goals of others by observing their actions over time. Remarkably, we can do so even when those actions lead to failure, enabling us to assist others when we detect that they might not achieve their goals. How might we endow machines with similar capabilities? Here we present an architecture capable of inferring an agent's goals online from both optimal and non-optimal sequences of actions. Our architecture models agents as boundedly-rational planners that interleave search with execution by replanning, thereby accounting for sub-optimal behavior. These models are specified as probabilistic programs, allowing us to represent and perform efficient Bayesian inference over an agent's goals and internal planning processes. To perform such inference, we develop Sequential Inverse Plan Search (SIPS), a sequential Monte Carlo algorithm that exploits the online replanning assumption of these models, limiting computation by incrementally extending inferred plans as new actions are observed. We present experiments showing that this modeling and inference architecture outperforms Bayesian inverse reinforcement learning baselines, accurately inferring goals from both optimal and non-optimal trajectories involving failure and back-tracking, while generalizing across domains with compositional structure and sparse rewards.