FallSafe-OVMM

Overview

Falls are a leading cause of injury and loss of independence for mobility-impaired individuals, and many occur during routine object retrieval. Although modern assistive robots can navigate, detect, and grasp objects effectively, current Open-Vocabulary Mobile Manipulation (OVMM) systems stop at task completion and do not evaluate whether the final placement of a retrieved object is safe or accessible for the user. To address this gap, we introduce FallSafe-OVMM, a safety-aware object placement framework that formulates placement as an optimization problem over human reachability and environmental risk. FallSafe-OVMM models the user through a Safe Pose Zone (SPZ), a parameterized reach envelope that represents seated and standing accessibility constraints. Given a scene in Habitat-Sim and a retrieved object, the system detects horizontal support surfaces and samples candidate placement locations proportional to surface area, filtering them through physical validity checks to ensure stable, collision-free configurations. Each candidate is evaluated using two interpretable geometric metrics: UnsafeReach, which captures reach difficulty based on horizontal distance, vertical deviation from ideal reach height, and reach envelope violations; and HazardExposure, which measures environmental risk factors including edge proximity, fall height, surface size, and clutter density. These components are combined into a unified SafetyScore, and the placement with minimum risk is selected. We evaluate FallSafe-OVMM across diverse simulated home environments drawn from HM3D, HSSD, and ReplicaCAD, and compare against heuristic baselines and a Vision-Language Model placement policy. Our geometric framework produces placements that closely align with human-annotated ground truth while operating in real time, and substantially improves over conventional distance- or edge-based strategies. Analysis further shows that explicitly modeling both reach difficulty and environmental hazards yields stronger placement decisions than either component alone. By formalizing safety-aware placement as a measurable and reproducible benchmark, FallSafe-OVMM advances OVMM toward user-centered evaluation and brings assistive robotics closer to reducing fall risk and supporting safer independent living.

Competition history

• CSEF 2026 Computational Science (Senior Division) · Entry S-07-18

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Computational Sciences projects
• CSEF Computational Sciences projects
• Every CSEF 2026 project