Locally Persistent Exploration in Continuous Control Tasks with Sparse Rewards
arXiv:2012.13658
Abstract
A major challenge in reinforcement learning is the design of exploration strategies, especially for environments with sparse reward structures and continuous state and action spaces. Intuitively, if the reinforcement signal is very scarce, the agent should rely on some form of short-term memory in order to cover its environment efficiently. We propose a new exploration method, based on two intuitions: (1) the choice of the next exploratory action should depend not only on the (Markovian) state of the environment, but also on the agent's trajectory so far, and (2) the agent should utilize a measure of spread in the state space to avoid getting stuck in a small region. Our method leverages concepts often used in statistical physics to provide explanations for the behavior of simplified (polymer) chains in order to generate persistent (locally self-avoiding) trajectories in state space. We discuss the theoretical properties of locally self-avoiding walks and their ability to provide a kind of short-term memory through a decaying temporal correlation within the trajectory. We provide empirical evaluations of our approach in a simulated 2D navigation task, as well as higher-dimensional MuJoCo continuous control locomotion tasks with sparse rewards.
arXiv commentTo be published in ICML, 2021
References in corpus (14)
- Continuous control with deep reinforcement learning
- Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor
- Noisy Networks for Exploration
- VIME: Variational Information Maximizing Exploration
- Reinforcement Learning with Deep Energy-Based Policies
- PEGASUS: A Policy Search Method for Large MDPs and POMDPs
- Count-Based Exploration with Neural Density Models
- GEP-PG: Decoupling Exploration and Exploitation in Deep Reinforcement Learning Algorithms
- EX2: Exploration with Exemplar Models for Deep Reinforcement Learning
- Dealing with Sparse Rewards in Reinforcement Learning
- Minimax Regret Bounds for Reinforcement Learning
- Better Exploration with Optimistic Actor-Critic
- Learning to Explore with Meta-Policy Gradient
- Benchmarking Bonus-Based Exploration Methods on the Arcade Learning Environment