Agents
Moving Out: Physically-grounded Human-AI Collaboration
The article introduces "Moving Out," a benchmark designed for physically-grounded human-AI collaboration, addressing the limitations of existing benchmarks that often overlook physical constraints. It presents a novel method called BASS (Behavior Augmentation, Simulation, and Selection) to enhance embodied agents' adaptability to diverse human behaviors and physical attributes. The benchmark includes two challenges and human-human interaction data, demonstrating that BASS outperforms state-of-the-art models in collaboration tasks, which is crucial for practitioners developing more effective AI systems that operate in real-world environments.
human-ai collaborationembodied agents