Johns Hopkins University · Department of Computer Science · Data Science and AI Institute · Laboratory for Computational Sensing and Robotics · Institute for Assured Autonomy

SpiderNoid: Crawling Control for a Humanoid

Overview

We enable a Unitree G1 humanoid (29-DoF) to transition from standing into a four-limbed crawling posture, crawl forward and turn, and return to standing. Two policies, one for the stand ↔ crawl transition and one for crawling (with two gait styles), are trained in mjlab with PPO and switched by a state machine with safety gates at each hand-over. The forearms serve as front legs: 3D-printed wrist braces lock the wrists and extend them into rounded feet.

Code, deployable policies, and hardware files are available on GitHub.

Contributors

Geeta Chandra Raju Bethala (demo) and Krishna Murthy Jatavallabhula (video)

Latest Posts

SpiderNoid: Crawling Control for a Humanoid

We enable a Unitree G1 humanoid to transition into a four-limbed crawling posture, crawl, and return to standing. The policies are trained in simulation and transfer directly to the real robot.

Turning Safety into Competence: Minimally Exploitable Robot Policies via Safety-Filtered RL

We propose a framework for training competitive robots that maintain safety without sacrificing performance, by separating safety synthesis from task learning in a two-stage reinforcement learning process. We formulate interactions as safety-critical games, prove that safety filtering preserves non-exploitability when players maintain safe maneuvers, and demonstrate superior performance over existing safe RL methods in quadruped touchdown games and hardware tests.

Winning a Won Game: Strict Reach-Avoid-Stay Control Barrier Functions

We study strict reach-avoid-stay (sRAS) control, which requires safely reaching a target and remaining there indefinitely after first entry. We present a Q-control barrier function filter for high-dimensional black-box systems with bounded uncertainty, merging a stay value encoding safe permanent residence with a reach-avoid value encoding safe target reachability. Our approach requires no knowledge of system dynamics, structural properties, or hand-crafted barriers, and we validate it on quadruped gap-jumping experiments (simulated and physical) and simulated autonomous racing scenarios.