AIIC AI Intelligence Centre

SOURCE-LINKED INTELLIGENCE

OmniNav: Robust Long-Horizon Target Navigation in Dynamic Environments

arXiv · AI, language, vision and robotics · article · Sep 8, 2026 · UTC

Long-horizon target navigation requires a robot to sustain task execution across evolving observations, decisions, and physical interactions. This requires three coupled capabilities: maintaining valid scene memory, revising target beliefs under partial observability, and selecting interaction-feasible navigation endpoints. However, the state underlying each capability is only conditionally valid: scene representations become stale when objects move or disappear, unsuccessful searches alter beliefs over target locations, and geometrically convenient endpoints may still be infeasible for manipu

Read original source ↗ Open in workspace

recordType
paper
region
Global

Evidence & attribution

First collected: 2026-09-20T20:22:01.598Z. This is not the publication date.