Looped Actor: Depth-Recurrent Reasoning Models for Reinforcement Learning

2026-09-29

AI Summary

This study addresses the limitation of fixed-computation policies in reinforcement learning, which struggle to adapt to varying task complexities. To this end, it proposes the Looped Actor model, introducing a Transformer-based deep looped policy network. By leveraging a parameter-shared recurrent architecture, the model enables input-dependent dynamic reasoning and adaptive computation allocation, supported by corresponding theoretical guarantees. The proposed approach is compatible with both online and offline reinforcement learning algorithms. Evaluated across 22 benchmark tasks, the Looped Actor achieves baseline performance levels using only one-sixteenth of the parameters while significantly enhancing multi-step planning capabilities and effectively reducing computational overhead.

Abstract

Looped reasoning models repeatedly apply a shared set of parameters, enabling more computation without increasing the model size. These models also support input-dependent computation by dynamically deciding when to stop looping. Motivated by the recent success of looped transformers in language modeling and reasoning, we investigate whether dynamic looping can similarly benefit sequential decision-making. We provide a complexity-theoretic motivation for this approach by showing that there exist Markov decision processes in which a state-adaptive policy achieves the optimal return with asymptotically less expected computation than any optimal fixed-runtime policy. To learn compute-adaptive policies in practice, we introduce Looped Actor, a transformer-based policy that repeatedly refines a latent representation toward a fixed point using a shared computational block. This allows the model to allocate computation adaptively by varying the number of loops based on the current state. We evaluate Looped Actor on 22 tasks across six environments, ranging from combinatorial puzzles to robotic manipulation and spanning online and offline reinforcement learning (RL) with discrete and continuous actions. Looped Actor matches or exceeds the performance of an untied baseline with 16$\times$ more parameters, with the largest gains in environments where action selection requires substantial multistep planning. For the Boxoban environment, we find that the computation allocation is structured: the number of loops increases with the number of remaining pushes and future optimal pushes become increasingly predictable from the latent state over successive loops. Together, these results highlight actor looping as a simple and efficient way to equip RL agents with adaptive computation and improve their planning capabilities. Code is available at https://github.com/camail-official/LoopedActor

Problem

• Reinforcement Learning
• Adaptive Computation
• Looped Reasoning
• Sequential Decision-Making
• Planning

Innovation

• Looped Actor
• Reinforcement Learning
• Adaptive Computation
• Depth-Recurrent Reasoning
• Weight Sharing