Embodied AI and Interactive Robotics

The field of embodied AI and interactive robotics is moving towards more realistic and dynamic simulations, with a focus on developing agents that can perform complex tasks in real-world environments. Recent developments have highlighted the importance of contingency-aware planning, interactive safety, and robustness in household environments. Noteworthy papers in this area include DualTHOR, which introduces a dual-arm humanoid simulation platform, and IS-Bench, which presents a benchmark for evaluating interactive safety in embodied agents. Other notable contributions include Judo, a user-friendly package for sampling-based model predictive control, and Mobile-R1, which employs interactive multi-turn reinforcement learning for mobile agents. These advancements are pushing the boundaries of what is possible in embodied AI and interactive robotics, and are likely to have significant impacts on the development of more capable and safe agents in the future. Notable papers: DualTHOR introduces a physics-based simulation platform for complex dual-arm humanoid robots. IS-Bench presents a benchmark for evaluating interactive safety in embodied agents, highlighting critical limitations in current agents.

Sources

DualTHOR: A Dual-Arm Humanoid Simulation Platform for Contingency-Aware Planning

IS-Bench: Evaluating Interactive Safety of VLM-Driven Embodied Agents in Daily Household Tasks

Judo: A User-Friendly Open-Source Package for Sampling-Based Model Predictive Control

Mobile-R1: Towards Interactive Reinforcement Learning for VLM-Based Mobile Agent via Task-Level Rewards

UAIbot: Beginner-friendly web-based simulator for interactive robotics learning and research

Agent-RewardBench: Towards a Unified Benchmark for Reward Modeling across Perception, Planning, and Safety in Real-World Multimodal Agents

Built with on top of