Time-Efficient Reinforcement Learning with Stochastic Stateful Policies
This addresses a key bottleneck in reinforcement learning for partially observable or complex environments, offering a practical improvement over existing methods.
The paper tackles the problem of slow and biased training of stateful policies in reinforcement learning due to Backpropagation Through Time (BPTT), by introducing a novel gradient estimator that decomposes policies into stochastic internal states and stateless components, resulting in faster and scalable performance on complex tasks like humanoid locomotion.
Stateful policies play an important role in reinforcement learning, such as handling partially observable environments, enhancing robustness, or imposing an inductive bias directly into the policy structure. The conventional method for training stateful policies is Backpropagation Through Time (BPTT), which comes with significant drawbacks, such as slow training due to sequential gradient propagation and the occurrence of vanishing or exploding gradients. The gradient is often truncated to address these issues, resulting in a biased policy update. We present a novel approach for training stateful policies by decomposing the latter into a stochastic internal state kernel and a stateless policy, jointly optimized by following the stateful policy gradient. We introduce different versions of the stateful policy gradient theorem, enabling us to easily instantiate stateful variants of popular reinforcement learning and imitation learning algorithms. Furthermore, we provide a theoretical analysis of our new gradient estimator and compare it with BPTT. We evaluate our approach on complex continuous control tasks, e.g., humanoid locomotion, and demonstrate that our gradient estimator scales effectively with task complexity while offering a faster and simpler alternative to BPTT.