Cross Anything: General Quadruped Robot Navigation through Complex Terrains
Shaoting Zhu, Derun Li, Yong Liu, Ningyi Xu, Hang Zhao
2024-07-24
Summary
This paper introduces the Cross Anything System (CAS), a new approach for helping quadruped robots navigate complex terrains using advanced reasoning and control techniques. It combines high-level planning with low-level movement control to improve how these robots can move in challenging environments.
What's the problem?
While robots have made progress in navigating their surroundings, many existing systems struggle with complex terrains like stairs, ramps, or uneven ground. Traditional methods often lack the ability to effectively plan routes and execute movements that adapt to these challenges, making it hard for robots to reach their goals safely and efficiently.
What's the solution?
CAS addresses these issues by using a two-part system: a high-level reasoning module that plans the robot's path and a low-level control policy that manages its movements. The high-level module uses a vision-language model to break down tasks into smaller steps and ensures the robot can adapt to changes in its environment. The low-level control uses a method called Probability Annealing Selection (PAS) to train the robot on how to move effectively across different terrains. This combination allows the robot to navigate complex 3D spaces while maintaining stability and accuracy.
Why it matters?
This research is important because it enhances the capabilities of quadruped robots, making them more effective in real-world applications such as search and rescue missions, delivery services, or exploration in difficult terrains. By improving how robots navigate complex environments, CAS can lead to safer and more reliable robotic systems that can assist humans in various tasks.
Abstract
The application of vision-language models (VLMs) has achieved impressive success in various robotics tasks, but there are few explorations for foundation models used in quadruped robot navigation. We introduce Cross Anything System (CAS), an innovative system composed of a high-level reasoning module and a low-level control policy, enabling the robot to navigate across complex 3D terrains and reach the goal position. For high-level reasoning and motion planning, we propose a novel algorithmic system taking advantage of a VLM, with a design of task decomposition and a closed-loop sub-task execution mechanism. For low-level locomotion control, we utilize the Probability Annealing Selection (PAS) method to train a control policy by reinforcement learning. Numerous experiments show that our whole system can accurately and robustly navigate across complex 3D terrains, and its strong generalization ability ensures the applications in diverse indoor and outdoor scenarios and terrains. Project page: https://cross-anything.github.io/