High-Dimensional Continuous Control Using Generalized Advantage Estimation

TLDR

Policy gradient methods are attractive in reinforcement learning because they directly optimize cumulative reward and can use nonlinear function approximators, yet they require many samples and suffer unstable improvement; moreover, our neural‑network policies map raw kinematics to joint torques without hand‑crafted representations. The study aims to reduce policy‑gradient variance via value functions and an exponentially weighted advantage estimator analogous to TD(λ), and to stabilize learning with trust‑region optimization applied to both policy and value neural networks. The authors employ value functions and an exponentially weighted advantage estimator to cut variance, and use trust‑region optimization for both the policy and value neural networks. The resulting algorithm achieves strong empirical performance on challenging 3D locomotion tasks, learning running gaits for bipedal and quadrupedal robots, enabling a biped to stand up from lying, and requires only 1–2 weeks of simulated experience for 3D biped tasks.

Abstract

Abstract: Policy gradient methods are an appealing approach in reinforcement learning because they directly optimize the cumulative reward and can straightforwardly be used with nonlinear function approximators such as neural networks. The two main challenges are the large number of samples typically required, and the difficulty of obtaining stable and steady improvement despite the nonstationarity of the incoming data. We address the first challenge by using value functions to substantially reduce the variance of policy gradient estimates at the cost of some bias, with an exponentially-weighted estimator of the advantage function that is analogous to TD(lambda). We address the second challenge by using trust region optimization procedure for both the policy and the value function, which are represented by neural networks. Our approach yields strong empirical results on highly challenging 3D locomotion tasks, learning running gaits for bipedal and quadrupedal simulated robots, and learning a policy for getting the biped to stand up from starting out lying on the ground. In contrast to a body of prior work that uses hand-crafted policy representations, our neural network policies map directly from raw kinematics to joint torques. Our algorithm is fully model-free, and the amount of simulated experience required for the learning tasks on 3D bipeds corresponds to 1-2 weeks of real time.