High-Dimensional Continuous Control Using Generalized Advantage Estimation

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(λ). 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.

Reinforcement Learning with Adaptive Curriculum Dynamics Randomization for Fault-Tolerant Robot Control

“…In the analysis, we introduce a slightly modified reward function instead of using the commonly applied function described in [24], because the existing reward function [24] is likely to result in a conservative policy in which the robot does not walk but remains in place. The existing reward function [24] can be defined by…”

On Effective Scheduling of Model-based Reinforcement Learning

“…There are multiple choices of the advantage function [23], and we use the baseline version of the Monte-Carlo returns to reduce the variance: Evaluate MBPO and construct the reward as the average return: R i = Avg(η);…”

Polymatrix Competitive Gradient Descent

“…This formulation allows us to use advanced, actor-critic-type approaches [25] to improve the sample efficiency. In our implementation, we use generalized advantage estimation (GAE) [26] A(s, a 1 , ..., a n ) = Q(s, a 1 , ..., a n ) − V (s) in place of the Q function to calculate the gradient and Hessian terms. After sampling a batch of states, actions, and rewards from the replay buffer, we construct two pseudo-objectives, one for the first derivative terms and one for the mixed Hessian terms required for the PCGD update.…”