Handling Delay in Real-Time Reinforcement Learning
Ivan Anokhin, Rishav Rishav, Matthew Riemer, Stephen Chung, Irina Rish, Samira Ebrahimi Kahou
Abstract
Real-time reinforcement learning (RL) introduces several challenges. First, policies are constrained to a fixed number of actions per second due to hardware limitations. Second, the environment may change while the network is still computing an action, leading to observational delay. The first issue can partly be addressed with pipelining, leading to higher throughput and potentially better policies. However, the second issue remains: if each neuron operates in parallel with an execution time of τ , an N -layer feed-forward network experiences observation delay of τ N . Reducing the number of layers can decrease this delay, but at the cost of the network's expressivity. In this work, we explore the trade-off between minimizing delay and network's expressivity. We present a theoretically motivated solution that leverages temporal skip connections combined with history-augmented observations. We evaluate several architectures and show that those incorporating temporal skip connections achieve strong performance across various neuron execution times, reinforcement learning algorithms, and environments, including four Mujoco tasks and all MinAtar games. Moreover, we demonstrate parallel neuron computation can accelerate inference by 6-350% on standard hardware. Our investigation into temporal skip connections and parallel computations paves the way for more efficient RL agents in real-time setting.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
Builds on8
- Mixtures of Experts Unlock Parameter Scaling for Deep RLJohan S. Obando-Ceron, Ghada Sokar, Timon Willi, Clare Lyle et al.ICML 2024 · 74 citations
- Improving Anytime Prediction with Parallel Cascaded Networks and a Temporal-Difference LossMichael L. Iuzzolino, Michael C. Mozer, Samy BengioNeurIPS 2021 · 15 citations
- Asynchronous Coagent NetworksJames E. Kostas, Chris Nota, Philip S. ThomasICML 2020 · 9 citations
- Learning by Competition of Self-Interested Reinforcement Learning AgentsStephen ChungAAAI 2022 · 5 citations
- Reinforcement Learning with Random DelaysYann Bouteiller, Simon Ramstedt, Giovanni Beltrame, Christopher J. Pal et al.ICLR 2021 · 3 citations
Related papers
- TempoRL: Learning When to ActAndré Biedenkapp, Raghu Rajan, Frank Hutter, Marius LindauerICML 2021 · 38 citations
- Enabling Realtime Reinforcement Learning at Scale with Staggered Asynchronous InferenceMatthew Riemer, Gopeshh Subbaraj, Glen Berseth, Irina RishICLR 2025
- Gradient Forward-Propagation for Large-Scale Temporal Video ModellingMateusz Malinowski, Dimitrios Vytiniotis, Grzegorz Swirszcz, Viorica Patraucean et al.CVPR 2021
- Variational Delayed Policy OptimizationQingyuan Wu, Simon Sinong Zhan, Yixuan Wang, Yuhui Wang et al.NeurIPS 2024 · 10 citations
- Autoregressive Dynamics Models for Offline Policy Evaluation and OptimizationMichael R. Zhang, Thomas Paine, Ofir Nachum, Cosmin Paduraru et al.ICLR 2021 · 52 citations
