Inhabitant targets long-horizon household interactions with spatial control
Yao Zhang, Zhuchenyang Liu, and Yu Xiao submitted Inhabitant to arXiv on November 18, 2026. The training-free agent framework specifies positions, paths, and poses, measures deviations from intended interactions, and revises its controls, while the accompanying Inhabit benchmark covers 75 household tasks across 31 apartment layouts. The authors report 98.7% task completion with Claude Opus 5.5, 93.3% with Claude Sonnet 5.5, and 82.7% with Gemini 4 Argon.
Inhabitant offers a concrete approach to long-horizon embodied interaction, pairing explicit spatial control with quantitative feedback in a new household-task benchmark.
Computer Science > Robotics
arXiv:2611.25005 (cs)
[Submitted on 18 Nov 2026]
Title:Inhabitant: An Agent Framework for Long-Horizon Human-Scene Interaction with Spatial Control and Quantitative Feedback
Authors:Yao Zhang, Zhuchenyang Liu, Yu Xiao
View a PDF of the paper titled Inhabitant: An Agent Framework for Long-Horizon Human-Scene Interaction with Spatial Control and Quantitative Feedback, by Yao Zhang and 2 other authors
#Inhabitant#Inhabit#Claude Opus 5.5#Claude Sonnet 5.5#Gemini 4 Argon