Learning Sidewalk Autopilot from Multi-Scale Imitation with Corrective Behavior Expansion

TL;DR

This paper introduces a novel imitation learning framework for sidewalk micromobility that enhances robustness and generalization by incorporating corrective behaviors and multi-scale hierarchical modeling, validated through real-world experiments.

Contribution

It proposes a multi-scale imitation learning architecture combined with corrective behavior expansion to improve control policy robustness in complex urban environments.

Findings

Significantly improved robustness in diverse sidewalk scenarios

Enhanced generalization through multi-scale hierarchical modeling

Effective recovery from mistakes demonstrated in real-world tests

Abstract

Sidewalk micromobility is a promising solution for last-mile transportation, but current learning-based control methods struggle in complex urban environments. Imitation learning (IL) learns policies from human demonstrations, yet its reliance on fixed offline data often leads to compounding errors, limited robustness, and poor generalization. To address these challenges, we propose a framework that advances IL through corrective behavior expansion and multi-scale imitation learning. On the data side, we augment teleoperation datasets with diverse corrective behaviors and sensor augmentations to enable the policy to learn to recover from its own mistakes. On the model side, we introduce a multi-scale IL architecture that captures both short-horizon interactive behaviors and long-horizon goal-directed intentions via horizon-based trajectory clustering and hierarchical supervision. Real-world experiments show that our approach significantly improves robustness and generalization in diverse sidewalk scenarios.