Robust Visual Teach-and-Repeat Navigation with Flexible Topo-metric Graph Map Representation

TL;DR

The paper tackles robust Visual Teach-and-Repeat navigation in dynamic, real-world environments by introducing a flexible topo-metric map that combines topological structure with local metric relations. It proposes keyframe clustering to improve loop detection, map expansion to incorporate novel observations, and a long-term, multi-goal tracking strategy coupled with a map-less local planner to maintain navigation when frame-map matching is imperfect. The approach is validated on a mobile platform, showing improved robustness over baselines, especially under obstacle occlusion and environmental changes, and demonstrates benefits from clustering and map expansion for loop detection and adaptation. The work advances practical VTR deployment by reducing global mapping requirements while enhancing reliable goal tracking and local trajectory optimization, with potential for end-to-end deep learning extensions in the future.

Abstract

Visual Teach-and-Repeat Navigation is a direct solution for mobile robot to be deployed in unknown environments. However, robust trajectory repeat navigation still remains challenged due to environmental changing and dynamic objects. In this paper, we propose a novel visual teach-and-repeat navigation system, which consists of a flexible map representation, robust map matching and a map-less local navigation module. During the teaching process, the recorded keyframes are formulated as a topo-metric graph and each node can be further extended to save new observations. Such representation also alleviates the requirement of globally consistent mapping. To enhance the place recognition performance during repeating process, instead of using frame-to-frame matching, we firstly implement keyframe clustering to aggregate similar connected keyframes into local map and perform place recognition based on visual frame-tolocal map matching strategy. To promote the local goal persistent tracking performance, a long-term goal management algorithm is constructed, which can avoid the robot getting lost due to environmental changes or obstacle occlusion. To achieve the goal without map, a local trajectory-control candidate optimization algorithm is proposed. Extensively experiments are conducted on our mobile platform. The results demonstrate that our system is superior to the baselines in terms of robustness and effectiveness.

Figures (14)

• Figure 1: Our system flowchart.
• Figure 2: Our keyframe clustering illustration. We cluster keyframes that are similar measured by DBoW2 tools. The keyframes to be erased project its feature points into the keyframe to be preserved. The preserved keyframe thus contains novel information from neighboring keyframes.
• Figure 3: Our motion planning illustration. We sample several angular velocity values and each corresponds to one candidate trajectory. To increase the covered area, we can repeat the candidate trajectory generation process at each endpoint of each generated trajectory. During the motion planning process, we compute the score of each trajectory at the first group. Angular velocity corresponding to the trajectory with the highest score is sent to the mobile platform.
• Figure 4: Our mobile platform. An IMU-embedded stereo camera and 3D LiDAR sensor are mounted for environmental perception and localization.
• Figure 5: Typical scenes of our office for teaching route collection.