TSD: A Physics-Inspired Trajectory Saliency Detector for Efficient Imitation Learning
2026-06-22 • Robotics
Robotics
AI summaryⓘ
The authors tackle the problem of expensive and limited high-quality data for teaching robots how to manipulate objects. They noticed that not all robot movements are equally important, so they created the Trajectory Saliency Detector (TSD) to identify the key parts of the robot's motions that really matter for task success. TSD uses two simple physics-based measures to pick out these important movements without needing extra training. Using TSD, the authors developed ways to shrink datasets without losing performance and to collect data more efficiently. Their experiments show that robots trained this way can learn with less but better data, making robot learning cheaper and faster.
Imitation LearningRobotic ManipulationTrajectory SaliencySpatial EntropyCentripetal AccelerationDataset CompressionDataset ExpansionSimulationRobot LearningMotion Analysis
Authors
Yiming Zhao, Gongrui Ma, Qingkai Li, Mingguo Zhao
Abstract
For imitation learning in robotic manipulation, high data collection costs result in the scarcity of high quality data. In this paper, we leverage the inherent heterogeneity of trajectories to address this challenge. Based on our observations of manipulation tasks, we categorize motions into transitional, precise, and agile types, defining the latter two as trajectory saliency due to their criticality to task success in contrast to the prevalent but less relevant transitional motions. Therefore, we propose the Trajectory Saliency Detector (TSD), a training-free and plug-and-play framework to identify trajectory saliency. TSD employs two physically-grounded metrics: spatial entropy to capture fine-grained manipulation and centripetal acceleration to detect agile maneuvering. We further leverage TSD to develop a dataset compression method that reduces training costs and a dataset expansion strategy that improves data collection efficiency. Extensive experiments in both simulation and real-world settings demonstrate that models trained on TSD-condensed datasets achieve comparable or even superior performance with 25% less data on average. These results validate the effectiveness of our dataset compression and expansion strategies, thereby confirming the utility of TSD. Consequently, TSD offers a scalable and cost-effective pathway to synthesize information-dense datasets for efficient robot learning. Project page: https://trajectory-saliency-detector.github.io/trajectory-saliency-detector/