LT-Exosense: A Vision-centric Multi-session Mapping System for Lifelong Safe Navigation of Exoskeletons

TL;DR

LT-Exosense tackles the problem of safe, long-term navigation for self-balancing exoskeletons by building a persistent, change-aware map from multiple RGB-D sessions. It fuses session maps via visual place recognition and a multi-session factor graph optimization, detects scene changes with volumetric differencing, and converts maps to elevation representations with traversability estimates to support global planning using a PRM. The work demonstrates accurate multi-session reconstruction (average point-to-point error below $5 ext{ cm}$ against ground-truth scans) and effective adaptive path planning in dynamic indoor environments, validating the approach on human and exoskeleton-carrying platforms. This framework paves the way for lifelong autonomy and safer operation of exoskeletons in real-world settings, with potential deployment in homes, rehabilitation facilities, and public spaces.

Abstract

Self-balancing exoskeletons offer a promising mobility solution for individuals with lower-limb disabilities. For reliable long-term operation, these exoskeletons require a perception system that is effective in changing environments. In this work, we introduce LT-Exosense, a vision-centric, multi-session mapping system designed to support long-term (semi)-autonomous navigation for exoskeleton users. LT-Exosense extends single-session mapping capabilities by incrementally fusing spatial knowledge across multiple sessions, detecting environmental changes, and updating a persistent global map. This representation enables intelligent path planning, which can adapt to newly observed obstacles and can recover previous routes when obstructions are removed. We validate LT-Exosense through several real-world experiments, demonstrating a scalable multi-session map that achieves an average point-to-point error below 5 cm when compared to ground-truth laser scans. We also illustrate the potential application of adaptive path planning in dynamically changing indoor environments.

Figures (3)

• Figure 1: Overview of the LT-Exosense system. Multiple SLAM sessions with keyframe images and point cloud submaps (a) are registered into one common reference frame, where change detection is performed to reflect the latest environment changes for safe navigation of exoskeletons (b). The merged map can be further converted into elevation maps with traversability estimates (c). Obstacle-free walkable paths can be planned on the elevation maps (d).
• Figure 2: Multi-session mapping and change detection results sequences ${}^{s1}H_a$ and ${}^{s2}H_b$. Subfigures (a) and (b) show merged maps from sequences recorded on two different days where no environmental change has occurred within each day. Subfigure (c) shows the merged maps from (a) and (b), where detected inter-day object-level change is highlighted in red and blue in the zoomed-in views with green arrows indicating the before and after changes. $\oplus$ here indicates sessions merging followed by change detection and map update.
• Figure 3: Change-aware adaptive path planning with LT-Exosense through progressive multi-session mapping. From left to right, each subfigure illustrates the planned path from the same start to the same goal as a new mapping session is incrementally merged. a) In Session 1, a direct path through the corridor is planned to reach the goal in the office. b) After merging Session 2, the map becomes more complete but two obstacles appear in the middle of the corridor. LT-Exosense re-plans the path to bypass the obstruction by rerouting through the meeting room. c) When the obstacle is removed in Session 3, the path planner can recover the shorter route again.