Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AI Olympics challenge with Evolutionary Soft Actor Critic

Marco Calì, Alberto Sinigaglia, Niccolò Turcato, Ruggero Carli, Gian Antonio Susto

TL;DR

A Model-free Deep Reinforcement Learning approach combined with an evolutionary strategy is proposed for the AI Olympics competition held at IROS 2024 based on a Model-free Deep Reinforcement Learning approach combined with an evolutionary strategy.

Abstract

In the following report, we describe the solution we propose for the AI Olympics competition held at IROS 2024. Our solution is based on a Model-free Deep Reinforcement Learning approach combined with an evolutionary strategy. We will briefly describe the algorithms that have been used and then provide details of the approach

AI Olympics challenge with Evolutionary Soft Actor Critic

TL;DR

A Model-free Deep Reinforcement Learning approach combined with an evolutionary strategy is proposed for the AI Olympics competition held at IROS 2024 based on a Model-free Deep Reinforcement Learning approach combined with an evolutionary strategy.

Abstract

In the following report, we describe the solution we propose for the AI Olympics competition held at IROS 2024. Our solution is based on a Model-free Deep Reinforcement Learning approach combined with an evolutionary strategy. We will briefly describe the algorithms that have been used and then provide details of the approach
Paper Structure (17 sections, 6 equations, 6 figures, 3 tables)

This paper contains 17 sections, 6 equations, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Competition's goal
  3. Soft Actor Critic
  4. Deep Deterministic Policy Gradient (DDPG)
  5. Soft Actor-Critic (SAC)
  6. Zero-Order Optimization and Evolutionary Strategies
  7. Separable Natural Evolution Strategy (SNES)
  8. Surrogate reward function
  9. On the agent robustness
  10. SAC agent robustness
  11. SNES agent robustness
  12. Agent training
  13. Main agent training
  14. Evolutionary strategy agent
  15. Experiments
  16. ...and 2 more sections

Figures (6)

  • Figure 1: State and input evolution of the final controller for the acrobot task
  • Figure 2: State and input evolution of the final controller for the pendubot task
  • Figure 3: Robustness benchmark results for pendubot (left) and acrobot (right)
  • Figure 4: State and input evolution on the unperturbed pendubot system.
  • Figure 5: State and input evolution on the perturbed pendubot system.
  • ...and 1 more figures