ANCHOR: A Physically Grounded Closed-Loop Framework for Robust Home-Service Mobile Manipulation

Anonymous Authors
Anonymous Affiliation
Paper Code (Coming Soon) Video

Method Improvements Overview

ANCHOR Teaser Image

Comparison between a conventional open-loop pipeline and ANCHOR on two representative failure modes. Top row: without operability-aware alignment, the robot navigates to a geometrically reachable but kinematically infeasible pose; ANCHOR actively repositions the base into the operable region. Bottom row: under external disturbances, ANCHOR re-anchors the world state, detects the displaced object, and replans to complete the task.

Abstract

Open-vocabulary mobile manipulation in real domestic environments requires reliable long-horizon execution under open-set object references and frequent disturbances. In practice, many failures arise not from semantic misunderstanding, but from inconsistency between symbolic plans and the evolving physical world. To address this execution inconsistency, we present ANCHOR, a physically grounded closed-loop framework that aligns symbolic reasoning with verifiable physical state during execution. ANCHOR integrates three mechanisms: (i) physically anchored task planning, (ii) operability-aware base alignment, and (iii) minimum-responsible-layer hierarchical recovery. Across 60 real-robot trials, ANCHOR improves task success from 53.3% to 71.7% and achieves a 71.4% recovery rate under perturbations.

Framework

ANCHOR Method Overview

ANCHOR emphasizes deployability and robustness via continual state re-anchoring. Rather than deferring feasibility checks to execution, ANCHOR couples perception, planning, and control throughout the task.

  • Physically Anchored Task Planning (PATP): Derives symbolic states from real-time sensor observations and re-evaluates them after each action.
  • Operability-Aware Base Alignment: Refines navigation endpoints using manipulator reachability and local collision constraints.
  • Minimum-Responsible-Layer Recovery: Attributes failures across perception, base-arm coordination, and execution, escalating only when necessary.

Physically Anchored Task Planning

Physically Anchored Task Planning

The PATP pipeline enforces a state-consistency contract: every symbolic predicate must be backed by directly observable, verifiable geometric evidence at execution time. The system restricts the LLM to generating a well-formed PDDL problem skeleton, while all geometry-related constraints and predicate evaluations are handled by lightweight sensing modules.

Overview Video

Demos

Demo 1: In-place grasp and place.

Demo 2: Task execution with navigation.

Demo 3: Object is disturbed but remains in view; the system recovers and completes the task.

Demo 4: Object is moved out of view; the system re-searches, retrieves it, and then places it.

Qualitative Results

Qualitative Results

Qualitative real-robot trial illustrating closed-loop recovery and operability-aware alignment. (1) Target search; (2) Base alignment ensures manipulation feasibility; (3) L1 recovery triggers local re-grasp after an initial failure; (4-5) Successful placement.