Pushing Through Clutter With Movability Awareness of Blocking Obstacles

TL;DR

This work tackles Navigation Among Movable Obstacles (NAMO) by introducing SVG-MPPI, a framework that simultaneously reasons about obstacle movability and planning through a Semantic Visibility Graph (SVG) and a physics-informed Model Predictive Path Integral (MPPI) controller. Movability is treated continuously via obstacle mass, enabling passage nodes near movable obstacles and avoiding explicit obstacle relocation; MPPI then optimizes trajectories to minimize contact forces with the environment using real-time physics simulations. The approach outperforms binary movability baselines in simulation and real-world tests, achieving higher path/planning success, lower contact forces, and capable online replanning. The results suggest that movability-aware planning combined with physics-based control offers robust navigation in cluttered, dynamic environments with practical implications for service and domestic robotics.

Abstract

Navigation Among Movable Obstacles (NAMO) poses a challenge for traditional path-planning methods when obstacles block the path, requiring push actions to reach the goal. We propose a framework that enables movability-aware planning to overcome this challenge without relying on explicit obstacle placement. Our framework integrates a global Semantic Visibility Graph and a local Model Predictive Path Integral (SVG-MPPI) approach to efficiently sample rollouts, taking into account the continuous range of obstacle movability. A physics engine is adopted to simulate the interaction result of the rollouts with the environment, and generate trajectories that minimize contact force. In qualitative and quantitative experiments, SVG-MPPI outperforms the existing paradigm that uses only binary movability for planning, achieving higher success rates with reduced cumulative contact forces. Our code is available at: https://github.com/tud-amr/SVG-MPPI

Figures (14)

• Figure 1: An overview of the proposed SVG-MPPI architecture where the SVG provides a weighted graph with efficient node placement around movable obstacles along which a lowest-effort path can be found. The generated set of waypoints guides the MPPI control strategy to efficiently sample rollouts around movable obstacles. If during interaction an obstacle is considered non-movable, the movability estimation gets updated and the path is replanned. Snapshots of a real-world example are shown on the left where the red star indicates the goal location and the masses of the obstacles are (A): 25 kg, (B): 20 kg, (C): 5 kg.
• Figure 3: Visualization of passage node construction between two sets of obstacles: (1) Hexagon and square (top row), and (2) Triangle and pentagon (bottom row). Each stage displays the progression from initial shape plotting to the identification of nearest points, boundary construction, and final passage node generation.
• Figure : (a)
• Figure : (a)
• Figure : (a)