Probabilistic Recurrent Intention Switching Model
For researchers in inverse reinforcement learning and robotics, PRISM provides a scalable, exact EM solution for multi-intention IRL, enabling temporally coherent goal switching in complex, non-Markovian environments.
PRISM introduces a recurrent network to model intention switching in inverse reinforcement learning, achieving the highest held-out log-likelihood across gridworld, mouse labyrinth, and BridgeData V2 robotic manipulation tasks while recovering interpretable intentions without labels.
Inverse reinforcement learning (IRL) recovers reward functions from observed behavior, yet traditional methods assume a single stationary reward that cannot capture goal switching within an episode. Recent multi-intention IRL methods address this by segmenting trajectories, but model intention transitions as either a memoryless Markov chain or via manual state augmentation with a fixed history window. We propose the Probabilistic Recurrent Intention Switching Model (PRISM), which replaces both mechanisms with a lightweight recurrent network that maps observation history to a per-step intention distribution. We prove that the resulting EM objective decomposes exactly into independent per-intention reward subproblems, each solvable in closed form, yielding an $\mathcal{O}(nK)$ E-step with no variational approximation. We evaluate PRISM on a non-Markovian gridworld, a mouse labyrinth, and BridgeData~V2 robotic manipulation, the first large-scale robotic application of multi-intention IRL. Across all settings PRISM achieves the highest held-out log-likelihood while recovering nameable, temporally coherent intentions from unlabeled demonstrations, suggesting that discrete goal switching is present in both biological and artificial agents.