Learning a Vision-Based Footstep Planner for Hierarchical Walking Control

2025-08-08

AI Summary

Existing bipedal locomotion on unstructured terrain suffers from fragility and poor real-time performance due to reliance on proprioceptive feedback or hand-crafted vision pipelines. To address this, we propose a vision-driven hierarchical control framework: a high-level gait planner based on reinforcement learning operates on local elevation maps for terrain-adaptive decision-making; a low-level controller employs a reduced-order linear inverted pendulum model with angular momentum dynamics to construct a low-dimensional state representation, coupled with operational-space control to generate stable foot trajectories. The framework establishes an end-to-end visual perception–planning–control loop. Evaluated on the Cassie robot in both simulation and hardware experiments, it achieves robust locomotion over challenging terrains—including slopes, gravel, and stairs—with enhanced stability and real-time capability (planning frequency >30 Hz), demonstrating effectiveness and generalization across diverse environments.

Technology Category

• Intelligent Robots: Behavior Learning & Control
• Computer Vision: Low Level & Physics-based Vision
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Agentic search

Abstract

Bipedal robots demonstrate potential in navigating challenging terrains through dynamic ground contact. However, current frameworks often depend solely on proprioception or use manually designed visual pipelines, which are fragile in real-world settings and complicate real-time footstep planning in unstructured environments. To address this problem, we present a vision-based hierarchical control framework that integrates a reinforcement learning high-level footstep planner, which generates footstep commands based on a local elevation map, with a low-level Operational Space Controller that tracks the generated trajectories. We utilize the Angular Momentum Linear Inverted Pendulum model to construct a low-dimensional state representation to capture an informative encoding of the dynamics while reducing complexity. We evaluate our method across different terrain conditions using the underactuated bipedal robot Cassie and investigate the capabilities and challenges of our approach through simulation and hardware experiments.

Problem

• Develop vision-based footstep planner for bipedal robots
• Overcome fragility of manual visual pipelines in real-world
• Enable real-time footstep planning in unstructured environments

Innovation

• Vision-based hierarchical control framework
• Reinforcement learning footstep planner
• Angular Momentum Linear Inverted Pendulum model