跨越萬物：通過複雜地形的四足機器人導航

Cross Anything: General Quadruped Robot Navigation through Complex Terrains

July 23, 2024

作者: Shaoting Zhu, Derun Li, Yong Liu, Ningyi Xu, Hang Zhao

cs.AI

摘要

視覺語言模型（VLMs）的應用在各種機器人任務中取得了令人印象深刻的成功，但在四足機器人導航中使用基礎模型的探索卻很少。我們介紹了Cross Anything System（CAS），這是一個創新系統，由高層推理模組和低層控制策略組成，使機器人能夠穿越複雜的3D地形並達到目標位置。對於高層推理和運動規劃，我們提出了一個新穎的算法系統，利用VLM的優勢，設計了任務分解和閉環子任務執行機制。對於低層運動控制，我們利用概率退火選擇（PAS）方法通過強化學習來訓練控制策略。大量實驗表明，我們的整個系統能夠準確且穩健地穿越複雜的3D地形，其強大的泛化能力確保了在各種室內和室外場景以及地形中的應用。項目頁面：https://cross-anything.github.io/

English

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/