Follow Everything: Goal-Aware Adaptation and Graph-Based Planner for Arbitrary Leader Following
Qianyi Zhang, Shijian Ma, Boyi Liu, Jingtai Liu, Jianhao Jiao, Dimitrios Kanoulas
AI summary
Problem
Existing methods fail to generalize to arbitrary leaders and struggle to re-identify or efficiently follow leaders when they temporarily leave the robot's field of view or when dynamic interactions occur.
Approach
The system uses a segmentation model with a distance frame buffer to track arbitrary leaders, a goal-aware adaptation mechanism to dynamically switch planning states based on visibility and motion, and a graph-based planner to generate time-optimal, obstacle-avoiding trajectories.
Key results
- 75.1% follow success rate in simulations and real-world tests
- 13.1% reduction in visual loss through distance-aware feature buffering
- 65.1% decrease in collisions via dynamic state adaptation
- 0.4 m average following distance maintained
Why it matters
Enables reliable, generalizable leader-following for assistive and exploration robots operating in dynamic, unstructured environments.
Abstract
Enabling robots to robustly follow leaders sup- ports tasks such as carrying supplies or guiding customers. While existing methods often fail to generalize to arbitrary leaders, and struggle when the leader temporarily leaves the robot’s field of view, this work presents a unified framework to address both challenges. First, a segmentation model replaces traditional category-specific detection models, allowing the leader to be of any shape or type. To improve robustness, a distance frame buffer is designed to store high-confidence leader embeddings across distance intervals, accounting for the unique characteristics of leader-following tasks. Second, a goal-aware adaptation mechanism is designed to govern robot planning states based on the leader’s visibility and motion, complemented by a graph-based planner that generates candidate trajectories for each state, ensuring efficient following with obstacle avoid- ance. Simulations and real-world experiments with a legged robot follower and diverse leaders in indoor and outdoor settings demonstrate an improved follow success rate of 75.1%, a reduced visual loss of 13.1%, a fewer collisions of 65.1%, and a shorter average distance of 0.4 m. Visit https://follow- everything.github.io for the video and code.