Learning to Run challenge solutions: Adapting reinforcement learning methods for neuromusculoskeletal environments

TL;DR

This paper analyzes eight deep reinforcement learning solutions from the NIPS 2017 Learning to Run challenge, highlighting their methods, modifications, and heuristics for controlling a musculoskeletal model to run efficiently.

Contribution

It provides a comparative overview of various RL approaches and modifications applied to neuromusculoskeletal control tasks in a competitive setting.

Findings

Multiple RL algorithms successfully addressed the challenge.

Common heuristics like reward shaping and action discretization were widely used.

Different teams implemented unique modifications of known algorithms.

Abstract

In the NIPS 2017 Learning to Run challenge, participants were tasked with building a controller for a musculoskeletal model to make it run as fast as possible through an obstacle course. Top participants were invited to describe their algorithms. In this work, we present eight solutions that used deep reinforcement learning approaches, based on algorithms such as Deep Deterministic Policy Gradient, Proximal Policy Optimization, and Trust Region Policy Optimization. Many solutions use similar relaxations and heuristics, such as reward shaping, frame skipping, discretization of the action space, symmetry, and policy blending. However, each of the eight teams implemented different modifications of the known algorithms.