Learning Getting-Up Policies for Real-World Humanoid Robots

2025-02-17

AI Summary

This work addresses the challenge of enabling full-size humanoid robots to autonomously recover from falls in realistic environments—specifically, from diverse initial postures (supine/prone) and complex terrains (inclined surfaces, snow, soft grass). We propose a two-stage curriculum learning framework: Stage I optimizes upright motion trajectories under relaxed constraints; Stage II refines them with smoothness, low-velocity, and robustness constraints to yield deployable policies. Our approach integrates high-fidelity collision-aware geometric modeling, sparse-reward reinforcement learning, and sim-to-real transfer techniques. To our knowledge, this is the first demonstration of successful autonomous stand-up on multiple real-world terrains—including flat, deformable, slippery, and inclined surfaces—by the G1 humanoid robot. It represents the world’s first learning-based, multi-terrain, real-world autonomous stand-up achievement for full-size humanoids, significantly advancing field adaptability and practical utility.

Technology Category

• Intelligent Robots: Learning & Optimization for ROB
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Search and Optimization: Learning to Search

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Automatic fall recovery is a crucial prerequisite before humanoid robots can be reliably deployed. Hand-designing controllers for getting up is difficult because of the varied configurations a humanoid can end up in after a fall and the challenging terrains humanoid robots are expected to operate on. This paper develops a learning framework to produce controllers that enable humanoid robots to get up from varying configurations on varying terrains. Unlike previous successful applications of humanoid locomotion learning, the getting-up task involves complex contact patterns, which necessitates accurately modeling the collision geometry and sparser rewards. We address these challenges through a two-phase approach that follows a curriculum. The first stage focuses on discovering a good getting-up trajectory under minimal constraints on smoothness or speed / torque limits. The second stage then refines the discovered motions into deployable (i.e. smooth and slow) motions that are robust to variations in initial configuration and terrains. We find these innovations enable a real-world G1 humanoid robot to get up from two main situations that we considered: a) lying face up and b) lying face down, both tested on flat, deformable, slippery surfaces and slopes (e.g., sloppy grass and snowfield). To the best of our knowledge, this is the first successful demonstration of learned getting-up policies for human-sized humanoid robots in the real world. Project page: https://humanoid-getup.github.io/

Problem

• Learning policies for humanoid fall recovery
• Addressing complex contact patterns in getting-up
• Real-world deployment on varied terrains

Innovation

• Learning framework for humanoid recovery
• Two-phase curriculum-based approach
• Real-world deployment on varied terrains