Robotics research explores SO(3) action representations in deep reinforcement learning

A new research paper explores the complexities of representing SO(3) actions in deep reinforcement learning, particularly for robotic control tasks. The study systematically evaluates common representations like Euler angles, quaternions, and rotation matrices across three standard algorithms (PPO, SAC, TD3) to understand their impact on exploration, training stability, and optimization. The findings suggest that representing actions as tangent vectors in a local frame offers the most reliable results across different algorithms and reward structures. AI

IMPACT Provides guidelines for selecting and using rotation actions in robotics, potentially improving RL agent performance in orientation-based tasks.

RANK_REASON The cluster contains an academic paper detailing research findings. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Euler angles
• Martin Schück
• Proximal Policy Optimization
• quaternions
• reinforcement learning
• robotics
• SO(3)
• rotation matrices
• TD3

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →