DualTHOR: A Dual-Arm Humanoid Simulation Platform for Contingency-Aware Planning
Boyu Li, Siyuan He, Hang Xu, Haoqi Yuan, Yu Zang, Liwei Hu, Junpeng Yue, Zhenxiong Jiang, Pengbo Hu, Börje F. Karlsson, Yehui Tang, Zongqing Lu
2025-06-26
Summary
This paper talks about DualTHOR, a simulation platform made for training humanoid robots with two arms. It uses real-world robot models and physics to help the robots learn how to handle complex tasks in home environments.
What's the problem?
The problem is that most simulators only work with simple robots, like those with wheels or just one arm, and they don’t include the chance for things to go wrong during actions, which makes it hard for robots to work well in real life where unexpected things happen.
What's the solution?
The researchers developed DualTHOR by building on an existing simulator and adding features like dual-arm control, realistic physics, and a contingency system that simulates things like objects breaking or spilling, so robots can practice recovering from failures and improve their planning and coordination.
Why it matters?
This matters because it makes it possible to train robots more realistically, helping them become better at handling everyday tasks that require both arms and adjusting to surprises, which is important for creating robots that can work safely and efficiently in homes and other real-world places.
Abstract
A simulator named DualTHOR for training dual-arm humanoid robots integrates real-world assets and physics to enhance the robustness and generalization of Vision Language Models.