From Physics to Physical Intelligence

从物理到机器人物理智能

We view robotics as the study of physical intelligence, emerging from the interaction between physics, learning, and embodiment.
我们将机器人研究视为“物理智能”的研究,其本质来源于物理建模、学习方法与具身系统的相互作用。


Research Philosophy

Physics → Learning → Embodiment → Physical Intelligence

  • Physics defines structure(物理提供结构)
  • Learning enables adaptation(学习提供适应性)
  • Embodiment unlocks capability(具身决定能力边界)

👉 Together, they enable robots to operate reliably in the real world
👉 三者共同作用,使机器人能够在真实世界中稳定运行


Layer 1: Physics-Based Modeling & Control

基于物理的建模与控制

We develop principled models and control algorithms grounded in dynamics, contact modeling, and optimization.
我们基于动力学、接触与优化理论,构建具有物理一致性的机器人建模与控制方法。

Key Topics

  • Contact dynamics & hybrid systems(接触动力学与混杂系统)
  • Control through Contact(接触交互控制)
  • Multi-contact kinodynamic motion planning(多接触运动规划)
  • Physically consistent estimation(物理一致状态估计)

Recent Works


Layer 2: Learning for Physical Systems

面向物理系统的学习方法

We develop learning-based methods that enhance physical models, enabling robots to adapt and generalize beyond analytical modeling.
我们发展面向物理系统的学习方法,在模型基础上提升机器人对复杂环境的适应能力与泛化能力。

Key Topics

  • Data-driven and augmented control(数据驱动及增强控制)
  • Physics embedded learning(物理驱动的学习)
  • System identification & calibration(系统辨识与标定)
  • Sim-to-Real transfer(仿真到现实迁移)

Recent Works


Layer 3: Embodied Robotic Systems & Design

具身机器人系统与本体设计

We design robotic systems where intelligence emerges from the interaction between hardware, control, and environment.
我们通过软硬件协同设计,使机器人智能从本体、控制与环境的交互中涌现。

Key Topics

  • Legged & humanoid locomotion(腿式与类人运动)
  • Loco-manipulation(移动操作)
  • Multi-modal robots(多模态机器人)
  • Mechanism & actuation design(本体结构与驱动设计)

Recent Works (Custom-built Robots)


Final Goal: Physical Intelligence

最终目标:机器人物理智能

Our ultimate goal is to build robots that can reliably move and interact in the real world.
我们的目标是让机器人能够在真实复杂环境中稳定运动与高效操作。

This requires the tight integration of:

  • Physics(结构)
  • Learning(适应性)
  • Embodiment(能力边界)