欢迎来到

Colab

空天具身智能实验室

CoLab（可乐实验室）成立于2016年6月12日。我们当前聚焦真实环境中的具身智能，重点方向包括UAV、VLA与多模态视频理解。同时，我们也开展面向自主智能体的三维感知与卫星星座智能调度研究，致力于构建能够在空-天-地场景中感知、推理与执行的智能体。

立即探索

代表性工作

UAVClaw: Autonomous UAV Agent for Real-World Tasks

UAVClaw brings OpenClaw task planning to real-world UAV autonomy, enabling drones to understand goals and execute tasks in open environments.

UAV Embodied AI OpenClaw

Project

Towards Realistic Earth-Observation Constellation Scheduling: Benchmark and Methodology

Satellite Scheduling Benchmark

Paper Project Code

"Hi AirStar, Guide Me to the Badminton Court."

Embodied AI UAV Navigation MM2025 DEMO

Paper Project Code

UAV-Flow Colosseo: A Real-World Benchmark for Flying-on-a-World UAV Imitation Learning

UAV-Flow Colosseo: A Real-World Benchmark for Flying-on-a-Word UAV Imitation Learning

UAV Imitation Learning Benchmark

Paper Project Code

RoboCerebra: A Large-scale Benchmark for Long-horizon Robotic Manipulation Evaluation

Robotics Dataset Embodied AI NeurIPS 2025

Paper Project Code

Directions

Embodied AI × UAVs

We build aerial embodied agents that integrate vision-language understanding with autonomous policy planning, enabling UAVs to perceive, think, and act in dynamic environments. Our research spans from simulation to real-world deployment, advancing aerial robots capable of intelligent navigation and multi-agent collaboration.

Visit UAV Home Page

Embodied AI × Satellites

Launching satellite constellations into orbit is hard, but efficiently scheduling them to execute tasks is equally challenging. As constellations scale up, manual planning can no longer keep pace with mission demands. We develop AI-driven scheduling and planning methods for large-scale satellite constellations, tackling the NP-hard challenges of dynamic Earth observation task allocation under tight time windows and complex physical constraints.

Embodied AI × Manipulation

Our manipulation research explores learning-based methods for robotic grasping, dexterous hand control, and human-object interaction. We develop policies that enable robots to manipulate diverse objects and perform complex tasks in unstructured environments.

Embodied AI × Perception

Beyond simple "seeing", we develop agents that deeply understand environments across spatial, temporal, and semantic dimensions. Key tasks include open-vocabulary detection and segmentation, enabling models to recognize and reason about novel objects.