FlowEval: How Much Adaptation Does an Embodied Foundation Model Need at Deployment?
FlowEval: How Much Adaptation Does an Embodied Foundation Model Need at Deployment?
This paper introduces FlowEval, a training-free diagnostic tool that uses flow-matching objectives to assess how well pretrained embodied foundation models fit target deployments. Evaluated across four model families (Pi-0.5, MolmoAct2, GR00T, DreamZero) on real-world and simulated manipulation tasks, FlowEval quantifies predictive disagreement on successful demonstrations to estimate required adaptation data. Finetuning Pi-0.5 and GR00T on 17 RoboCasa365 tasks further demonstrates that FlowEval correlates with success rate improvements as additional adaptation data is provided.
The authors introduce FlowEval, a training-free diagnostic method using flow-matching objectives to estimate adaptation requirements for embodied foundation models across four model families (Pi-0.5, MolmoAct2, GR00T, DreamZero) on real-world and simulated manipulation tasks.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
arXiv:2611.11650v1 Announce Type: new Abstract: Large-scale pretraining is reducing the amount of adaptation required to deploy embodied foundation models, but it is not guaranteed that a pretrained model already provide the level of performance required in every deployment. Practitioners therefore need to understand how well a pretrained model matches a target deployment and how much adaptation data is left to be co
Open source S2
Source：arXiv Robotics — research abstracts · arxiv.org