Embodiment-Aware Inference-Time Search for Human Reaction Selection in Humanoid Retargeting

Key points

• Training humanoid robots to learn human behaviors requires expensive data collection, and human-robot interaction data can be even more difficult and costly to collect.1
• Training a specialized reaction model for each humanoid embodiment can also be time-consuming and resource-intensive.1
• Human-to-humanoid motion retargeting provides a more scalable alternative, allowing human-human reaction data to serve as a common motion representation for different humanoid embodiments.1
• Experimental results demonstrate that the proposed framework maintains overall human reaction quality while improving the feasibility of retargeted humanoid reactions across different embodiments.1

Sources (1)

• [1]Embodiment-Aware Inference-Time Search for Human Reaction Selection in Humanoid Retargeting
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 11, 02:09 AM

  In this work, we propose an inference-time reaction search framework that leverages stochastic sampling from a generative reaction model and jointly considers reaction quality and humanoid embodiment feasibility.

  Training humanoid robots to learn human behaviors requires expensive data collection, and human-robot interaction data can be even more difficult and costly to collect.

Extractive summary: sentences quoted from the sources.