Plasticity preserving methods improve rl robustness under noise
Towards Understanding the Impact of Plasticity Loss on Reinforcement Learning in Stochastic Environments
Machine Learning
Summary
Reinforcement learning algorithms can struggle to learn well when the environment is noisy. The authors studied how loss of adaptability, called plasticity loss, affects learning in noisy settings. They found that methods that keep neural networks adaptable help learning, especially as observation noise grows. Interestingly, noise can sometimes help by acting like a regularizer, but in tough environments, instability becomes a bigger problem than plasticity loss.
What this means in practice
- •For reinforcement learning developers: Use plasticity-preserving techniques to make agents learn more reliably in noisy control tasks.
- •For robotics engineers: Improve decision-making robustness of robotic systems operating in unpredictable or noisy conditions by managing plasticity loss.
Authors
Philipp Bordne, André Biedenkapp
Abstract
Deep reinforcement learning (RL) is a powerful framework for sequential decision-making. Despite many high-profile successes in recent years, deep RL is known to be brittle. An emerging line of research attributes this brittleness to a loss of plasticity in neural networks. However, plasticity loss has so far been understudied in stochastic environments. We study this interplay systematically, training SAC agents on three DeepMind Control tasks augmented with controlled levels of observation and action noise. We find that plasticity-preserving methods increase learning robustness on the two harder tasks, with their advantage growing as observation noise increases, while combining resets with layer normalization underperforms resets alone at every observation noise level tested. Further, stochasticity can accelerate the loss of plasticity in harder environments, whereas in easier settings noise can act as a regularizer that sustains plasticity. Lastly, plasticity loss is most pronounced in the early stages of training. In the most challenging stochastic settings, TD instability rather than gradual plasticity loss becomes the dominant failure mode.