aiqbench analysis
EgoContact is a benchmark dataset of 22.5K manually verified egocentric interaction events with pixel-level contact annotations, used to test whether learned models can predict where a robot or agent should touch an object. The authors report that current methods often miss precise contact points and sometimes fail to find the target object in varied real-world scenes. Their reported gains from using the dataset as a memory bank for an affordance-transfer policy (grasp success rising from 14.2% to 31.2% in simulation and from 16% to 58% on a real robot) are notable, but they come from one policy and need independent replication before readers treat them as general.
EgoContact: Evaluating Physical Intelligence with Egocentric Video
arXiv:2610.39167v1 Announce Type: new Abstract: Egocentric video has emerged as one of the most scalable sources of data for learning generalizable manipulation priors. Yet, despite its growing use in pretraining robotic policies, it has been underutilized as a validation tool for physical intelligence, a key component of which is understanding where to make contact with an object to achieve an intended interaction. We present EgoContact, a large-scale egocentric dataset for evaluating the generalizability and precision of learned manipulation priors through future contact-point prediction. EgoContact contains 22.5K manually verified interaction events, each with a pre-contact observation, semantic intent, and pixel-precise contact locations. We find that current methods often struggle to predict precise interaction contacts and sometimes fail to localize the target object in diverse, in-the-wild scenes, even when allowed multiple predictions. Furthermore, we demonstrate the usefulness of EgoContact's annotated interactions for robotic manipulation: simply using EgoContact as the memory bank of an existing affordance-transfer policy improves grasp success from 14.2% to 31.2% in simulation and from 16% to 58% on a real robot.
Abstract:Egocentric video has emerged as one of the most scalable sources of data for learning generalizable manipulation priors. Yet, despite its growing use in pretraining robotic policies, it has been underutilized as a validation tool for physical intelligence, a key component of which is understanding where to make contact with an object to achieve an intended interaction. We present EgoContact, a large-scale egocentric dataset for evaluating the generalizability and precision of learned manipulation priors through future contact-point prediction. EgoContact contains 22.5K manually verified interaction events, each with a pre-contact observation, semantic intent, and pixel-precise contact locations. We find that current methods often struggle to predict precise interaction contacts and sometimes fail to localize the target object in diverse, in-the-wild scenes, even when allowed multiple predictions. Furthermore, we demonstrate the usefulness of EgoContact's annotated interactions for robotic manipulation: simply using EgoContact as the memory bank of an existing affordance-transfer policy improves grasp success from 14.2% to 31.2% in simulation and from 16% to 58% on a real robot.
Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2610.39167 [cs.RO]
(or arXiv:2610.39167v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2610.39167 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 264 words