From Imitation to Alignment: Human-Preference Flow Policies for Long-Horizon Sidewalk Navigation

2026-06-10 · Source: Artificial Intelligence · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

FlowPilot is a novel mapless navigation policy designed for autonomous long-horizon sidewalk navigation, utilizing only a single monocular RGB camera. Developed to overcome limitations of traditional imitation learning, such as compounding errors and poor social compliance, FlowPilot employs anchored flow matching for policy pre-training on extensive robot fleet data. This approach captures complex, multimodal sidewalk navigation behaviors. To enhance counterfactual reasoning and social compliance, the policy integrates a human-in-the-loop preference learning scheme, fine-tuned with minimal human intervention data. Evaluated in diverse simulation and real-world environments, FlowPilot achieved a 42% success rate and 66% route completion in simulation. Its human-preference tuned variant, FlowPilot-HP, further demonstrated improved real-world robustness and social compliance, reducing IR by 40.0% and NIR by 52.1% compared to the base model, making it suitable for micro-mobility applications like robotic food delivery.

Key takeaway

For Robotics Engineers developing autonomous micro-mobility solutions, FlowPilot offers a robust approach to long-horizon sidewalk navigation. If your current imitation learning policies suffer from compounding errors or social compliance issues, consider integrating anchored flow matching for pre-training and a human-in-the-loop preference learning scheme. This method can significantly improve real-world robustness and social compliance, reducing intervention rates and enhancing safety for your robotic food delivery or assistive wheelchair applications.

Key insights

FlowPilot combines anchored flow matching with human-in-the-loop preference learning for robust, socially compliant sidewalk navigation.

Principles

• Imitation learning alone struggles with complex, social navigation.
• Human preference data improves counterfactual reasoning.
• Anchored flow matching captures diverse behaviors.

Method

Pre-train with anchored flow matching on robot fleet data, then fine-tune using human-in-the-loop preference learning with intervention data to enhance social compliance and counterfactual reasoning.

In practice

• Deploy FlowPilot for micro-mobility applications.
• Use human feedback to refine autonomous navigation.
• Leverage monocular RGB for lightweight perception.

Topics

• Sidewalk Navigation
• Autonomous Robotics
• Imitation Learning
• Preference Learning
• Flow Matching
• Micro-mobility
• Monocular Vision

Best for: Computer Vision Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• From Imitation to Alignment: Human-Preference Flow Policies for Long-Horizon Sidewalk Navigation — Bridging imitation learning with human preference alignment significantly enhances robot navigation robustness and social compliance.
• FlowMPC: Improving Flow Matching policies with World Models — Integrating world models with Flow Matching policies via planning significantly improves test-time performance in multimodal action spaces.
• SPADE: Sketch-guided Path Planning Augmented with Diffusion Experts — SPADE enhances AMR path planning via diffusion-augmented behavioral cloning, improving generalization and robustness with fewer parameters.
• PLAN-S: Bridging Planning with Latent Style Dynamics for Autonomous Driving World Models — PLAN-S enhances autonomous driving LWMs by explicitly modeling style-conditioned spatial costs for improved controllability and safety.
• Multi-Agent Reinforcement Learning for Safe Autonomous Driving Under Pedestrian Behavioral Uncertainty — Co-training SDCs with personality-driven MARL pedestrians significantly improves safety by exposing vehicles to realistic, uncertain human behaviors.
• Beyond Waypoints: A Trajectory-Centric Waypointing Paradigm for Vision-Language Navigation — Grounding waypoints in executable trajectories resolves planning-control inconsistencies and improves reachability in Vision-Language Navigation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.