Ratatouille: Imitation Learning Ingredients for Real-world Social Robot Navigation

2025/09/21 by Mithun Vanniasinghe, Han, James R., Hshmat Sahak +6 · 1 citation
Computer Science · Psychology · #FOS: Computer and information sciences #Multimodal Machine Learning Applications #Reinforcement Learning in Robotics #Robotics (cs.RO) #Social Robot Interaction and HRI

paper · pdf · doi:10.48550/arxiv.2509.17204

openalex publication_date 2025/09/21 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Scaling Reinforcement Learning to in-the-wild social robot navigation is both data-intensive and unsafe, since policies must learn through direct interaction and inevitably encounter collisions. Offline Imitation learning (IL) avoids these risks by collecting expert demonstrations safely, training entirely offline, and deploying policies zero-shot. However, we find that naively applying Behaviour Cloning (BC) to social navigation is insufficient; achieving strong performance requires careful architectural and training choices. We present Ratatouille, a pipeline and model architecture that, without changing the data, reduces collisions per meter by 6 times and improves success rate by 3 times compared to naive BC. We validate our approach in both simulation and the real world, where we collected over 11 hours of data on a dense university campus. We further demonstrate qualitative results in a public food court. Our findings highlight that thoughtful IL design, rather than additional data, can substantially improve safety and reliability in real-world social navigation. Video: https://youtu.be/tOdLTXsaYLQ. Code will be released after acceptance.

Citations

• Reinforcement Learning with Action Chunking
• π_0.5: a Vision-Language-Action Model with Open-World Generalization
• The Pitfalls of Imitation Learning when Actions are Continuous
• Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success
• Multi-Transmotion: Pre-trained Model for Human Motion Prediction
• DR-MPC: Deep Residual Model Predictive Control for Real-world Social Navigation
• A Comprehensive Survey on Data Augmentation
• On the Foundations of Shortcut Learning
• SACSoN: Scalable Autonomous Control for Social Navigation
• Bigger, Better, Faster: Human-level Atari with human-level efficiency
• NaviSTAR: Socially Aware Robot Navigation with Hybrid Spatio-Temporal Graph Transformer and Preference Learning
• Toward Human-Like Social Robot Navigation: A Large-Scale, Multi-Modal, Social Human Navigation Dataset
• Diffusion Policy: Visuomotor Policy Learning via Action Diffusion
• Diffusion policy: Visuomotor policy learning via action diffusion
• Intention Aware Robot Crowd Navigation with Attention-Based Interaction Graph
• Fighting Copycat Agents in Behavioral Cloning from Observation Histories
• MMDetection: Open MMLab Detection Toolbox and Benchmark
• Causal Confusion in Imitation Learning
• DeepMoTIon: Learning to Navigate Like Humans
• Socially Compliant Navigation through Raw Depth Inputs with Generative\n Adversarial Imitation Learning

Cited by

• SONG: A Photorealistic 3D Gaussian Simulation Platform for Benchmarking Social Navigation

Related

• PIRLNav: Pretraining with Imitation and RL Finetuning for ObjectNav
• VITAL: Interactive Few-Shot Imitation Learning via Visual Human-in-the-Loop Corrections
• Robust Instant Policy: Leveraging Student's t-Regression Model for Robust In-context Imitation Learning of Robot Manipulation
• Self-Augmented Robot Trajectory: Efficient Imitation Learning via Safe Self-augmentation with Demonstrator-annotated Precision
• Integrating Offline Pre-Training with Online Fine-Tuning: A Reinforcement Learning Approach for Robot Social Navigation