ROBOFOUNDRY
arXiv · 2026
RoboFoundry: System-as-Policy Evolution for Self-Learning Embodied Agents
Evolving memory and skill systems from execution experience to build self-learning embodied agents across robots.
Learning to act in the physical world
Shuhao Liao
Ph.D. Student · Beihang University
I am a Ph.D. student at the State Key Laboratory of Complex and Critical Software Environment (SKLCCSE), Beihang University, advised by Wenjun Wu and Jie Luo. In 2025–2026, I am a joint Ph.D. student at the National University of Singapore (NUS), advised by Guillaume Sartoretti.
My research focuses on robot learning and embodied intelligence, spanning reinforcement learning, autonomous exploration, legged locomotion, and multi-agent decision-making. I am interested in connecting learning algorithms with reliable behavior on real robots.
Research interests
Reinforcement learningEmbodied intelligenceMulti-agent systems
News
- 1 paper on vision-based legged locomotion (KiVi) is accepted to IROS 2026.
- 1 paper on geometric multi-agent pathfinding (SIGMA) is accepted to ICRA 2025.
- 1 paper on partial symmetry in multi-agent reinforcement learning (PSE) is accepted to AAAI 2024.
- 1 paper on a many-agent reinforcement learning platform for aerial robots (Air-M) is accepted to IROS 2023.
Selected publications
8 works* Equal contribution
ROBOFOUNDRY
FARE
arXiv · 2026
FARE: Fast-Slow Agentic Robotic Exploration
Connecting language-based global reasoning with reactive local control for autonomous exploration.
GPO
arXiv · 2026
GPO: Growing Policy Optimization for Legged Robot Locomotion and Whole-Body Control
Progressively expanding the action space to make reinforcement learning for legged robots more effective.
TAGA
arXiv · 2026
TAGA: Terrain-aware Active Gaze Learning for Generalizable Agile Humanoid Locomotion
Learning to attend to informative terrain for agile, perceptive humanoid locomotion.
KIVI
IROS · 2026
KiVi: Kinesthetic-Visuospatial Integration for Dynamic and Safe Egocentric Legged Locomotion
Balancing proprioception and vision for robust quadruped locomotion under visual disturbances.
SIGMA
ICRA · 2025
SIGMA: Sheaf-Informed Geometric Multi-Agent Pathfinding
Using local geometric consensus to enable decentralized cooperation and collision-free pathfinding.
PSE
AAAI · 2024
Leveraging Partial Symmetry for Multi-Agent Reinforcement Learning
Adaptively exploiting partial symmetry to improve learning efficiency and multi-agent coordination.
AIR-M
IROS · 2023
Air-M: A Visual Reality Many-Agent Reinforcement Learning Platform for Large-Scale Aerial Unmanned System
A many-agent reinforcement learning platform for research on large-scale aerial robot systems.