Data-Asymmetric Latent Imagination and Reranking for 3D Robotic Imitation Learning

2026-05-11

AI Summary

This work addresses the challenge in real-world robotic imitation learning where high-quality demonstrations are scarce and abundant suboptimal or failed trajectories remain underutilized. To tackle this, the authors propose DALI-R, a novel framework that, for the first time in 3D imitation learning, integrates asymmetric data utilization, imagination replay via a 3D point cloud–based latent world model, and a task-completion scorer to re-rank action segments. This approach effectively extracts informative environmental dynamics and failure patterns from low-quality trajectories. By combining diffusion policies, flow matching, and action-chunk re-ranking, DALI-R achieves an average improvement of 6.8% in task success rate on the Adroit and MetaWorld benchmarks, with less than a 0.7× increase in inference overhead.

Abstract

Robotic imitation learning typically assumes access to optimal demonstrations, yet real-world data collection often yields suboptimal, exploratory, or even failed trajectories. Discarding such data wastes valuable information about environment dynamics and failure modes, which can instead be leveraged to improve decision-making. While 3D policies reduce reliance on high-quality demonstrations through strong spatial generalization, they still require large-scale data to achieve high task success. To address this, we propose DALI-R, a Data-Asymmetric Latent Imagination and Reranking framework for 3D robotic imitation learning from mixed-quality trajectories. It learns a Latent World Model over 3D point clouds for imagined rollouts and a Task Completion Scorer that reranks candidate action chunks, improving decision-making without additional high-quality demonstrations. We instantiate DALI-R with both diffusion and efficient flow-matching policies and evaluate it on Adroit and MetaWorld benchmarks. Across the two evaluated 3D base policies, DALI-R achieves an average $6.8$\% improvement in success rate while incurring less than $0.7\times$ additional inference overhead.

Problem

• Imitation Learning
• 3D Robotics
• Suboptimal Demonstrations
• Mixed-quality Trajectories
• Data Efficiency

Innovation

• Latent World Model
• Imitation Learning
• 3D Point Clouds
• Reranking
• Flow Matching