Summary
Choosing the best action for robots or self-driving cars is tricky because some methods rely on guessing small changes continuously, which can be hard to understand and might create unlikely actions. To fix this, the researchers developed Large Discrete Policy (LDiP), which picks actions from a big set of realistic possibilities instead of tweaking them a little bit. They improve decision-making by repeatedly scoring and narrowing down the choices in a way that balances careful ranking and randomness. Testing LDiP in tasks like driving and robot control showed it works better or just as well as older methods, making behavior more understandable and believable.
behavior policydiscrete policygenerative modelstochastic iterative scoringaction selectionrobotic manipulationautonomous drivingdecision process
Authors
Zhenxin Li, Nadine Chang, Xinglong Sun, Jingde Chen, Wenhao Yao, Zi Wang, Maying Shen, Yu-Gang Jiang, Zuxuan Wu, Shiyi Lan, Jose M. Alvarez
Abstract
Behavior policies are often formulated as continuous generative models, whose iterative denoising processes are expressive but difficult to interpret and prone to producing implausible actions. We propose the Large Discrete Policy (LDiP), a fully discrete behavior modeling framework that selects actions from a large vocabulary of physically plausible candidates. Rather than perturbing actions, LDiP improves expressivity through stochastic iterative scoring: it progressively re-scores and prunes candidates with score-space stochasticity, enabling fine-grained ranking and exploration among plausible actions while preserving an explicit decision process. Across end-to-end planning, closed-loop driving, robotic manipulation, and vision-language-action settings, LDiP consistently outperforms strong discrete and continuous baselines in autonomous driving, and exceeds or matches continuous generative policies in robotic manipulation. These results show that discrete policies, when equipped with effective scoring mechanisms, offer an expressive, plausible, and interpretable alternative for behavior modeling. Project website: https://zhenxinli.net/LargeDiscretePolicy/.