# Sihang Li and coauthors present FLORE for force-reactive humanoid loco-manipulation [Canonical reading](https://escapemeridian.com/readings/sihang-li-and-coauthors-present-flore-for-force-reactive-15a7d624) Sihang Li and coauthors present FLORE, a framework for humanoid robots to coordinate movement and contact force during tasks involving physical contact. Its force-reactive diffusion policy uses vision, proprioception, and measured wrenches to predict actions and per-hand force commands; a hybrid compliance whole-body controller executes them while maintaining balance. The authors report that, across three contact-rich robot tasks, the policy succeeded in 80% of rollouts, compared with 26.7% for a position-only policy trained on the same demonstrations and run with the same controller. They also report that, in simulation, the controller tracked force commands of varied magnitude and direction with little loss in whole-body motion accuracy. Escape Meridian’s reading: These results suggest a possible benefit of representing force as well as motion when a humanoid interacts with objects, but the abstract does not establish performance beyond the reported tasks. It does not identify the tasks or report how results vary across conditions. How well does FLORE perform on different contact-rich tasks or under changed conditions? Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source arXiv eess.SY · Article Original source date: Oct 26, 2026 (2026-10-26). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-04T23:02:06.691Z. First observed: 2026-10-27T04:17:11.818Z. Retrieved: 2026-10-29T22:33:20.443Z. - [Learning Force-Reactive Humanoid Loco-Manipulation](https://arxiv.org/abs/2610.35862) ## Limits The abstract does not name the three robot tasks or describe the conditions and variation across rollouts. The reported results cover the stated simulation and three contact-rich robot tasks; the abstract does not establish performance on other tasks or conditions. ## Open question How well does FLORE perform on different contact-rich tasks or under changed conditions? ## Source-stated The authors present FLORE as a framework for learning force-reactive humanoid loco-manipulation. > we present FLORE, a framework for learning force-reactive humanoid loco-manipulation. - [Supporting source](https://arxiv.org/abs/2610.35862) ## Source-stated The policy uses vision, proprioception, and measured wrenches to predict actions and per-hand force commands. > A force-reactive diffusion policy takes vision, proprioception, and measured wrenches as input, then predicts actions and per-hand force commands. - [Supporting source](https://arxiv.org/abs/2610.35862) ## Source-stated Across three contact-rich robot tasks, the authors report 80% success for the force-reactive policy and 26.7% for a position-only policy trained on the same demonstrations and executed by the same controller. > Across three contact-rich tasks on the robot, the force-reactive policy succeeds in 80% of rollouts, compared with 26.7% for a position-only policy trained on the same demonstrations and executed by the same controller. - [Supporting source](https://arxiv.org/abs/2610.35862) ## Source-stated In simulation, the authors report force tracking with little loss in whole-body motion accuracy. > In simulation, the controller tracks per-hand force commands of varied magnitude and direction, with little loss in whole-body motion accuracy. - [Supporting source](https://arxiv.org/abs/2610.35862) ## Escape Meridian interpretation The reported robot results suggest a possible benefit from using force-reactive control for the tested contact-rich tasks; the abstract does not establish performance on other tasks. > Across three contact-rich tasks on the robot, the force-reactive policy succeeds in 80% of rollouts - [Supporting source](https://arxiv.org/abs/2610.35862) ## Source and contributor context Li, Sihang, Guo, Xiaofeng, Cao, Jinkun, Cheng, Charlie, Yang, Ruihan, Feng, Chen, Shi, Guanya, Qi, Haozhi · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: 1afc0a31-2573-4d6c-8242-793b74d0c721. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-04T23:02:06.691Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 28a78eb94371ce53e1b0064626a93d1a5bf31bb2757e133b2821412a200bc8f2. Source text hash: bcf793aafcec07a74b75babf8f874c0e7b0e8d46b75dc39903c166c0afd417cd. ## Revision and source-check history ### Version 1 — 2026-11-04T23:02:06.691Z Sihang Li and coauthors present FLORE, a framework for humanoid robots to coordinate movement and contact force during tasks involving physical contact. Its force-reactive diffusion policy uses vision, proprioception, and measured wrenches to predict actions and per-hand force commands; a hybrid compliance whole-body controller executes them while maintaining balance. The authors report that, across three contact-rich robot tasks, the policy succeeded in 80% of rollouts, compared with 26.7% for a position-only policy trained on the same demonstrations and run with the same controller. They also report that, in simulation, the controller tracked force commands of varied magnitude and direction with little loss in whole-body motion accuracy. Escape Meridian’s reading: These results suggest a possible benefit of representing force as well as motion when a humanoid interacts with objects, but the abstract does not establish performance beyond the reported tasks. It does not identify the tasks or report how results vary across conditions. How well does FLORE perform on different contact-rich tasks or under changed conditions? ## Original source arXiv eess.SY · Article Original source date: Oct 26, 2026 (2026-10-26). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-04T23:02:06.691Z. First observed: 2026-10-27T04:17:11.818Z. Retrieved: 2026-10-29T22:33:20.443Z. - [Learning Force-Reactive Humanoid Loco-Manipulation](https://arxiv.org/abs/2610.35862) ## Limits The abstract does not name the three robot tasks or describe the conditions and variation across rollouts. The reported results cover the stated simulation and three contact-rich robot tasks; the abstract does not establish performance on other tasks or conditions. ## Open question How well does FLORE perform on different contact-rich tasks or under changed conditions? ## Source-stated The authors present FLORE as a framework for learning force-reactive humanoid loco-manipulation. > we present FLORE, a framework for learning force-reactive humanoid loco-manipulation. - [Supporting source](https://arxiv.org/abs/2610.35862) ## Source-stated The policy uses vision, proprioception, and measured wrenches to predict actions and per-hand force commands. > A force-reactive diffusion policy takes vision, proprioception, and measured wrenches as input, then predicts actions and per-hand force commands. - [Supporting source](https://arxiv.org/abs/2610.35862) ## Source-stated Across three contact-rich robot tasks, the authors report 80% success for the force-reactive policy and 26.7% for a position-only policy trained on the same demonstrations and executed by the same controller. > Across three contact-rich tasks on the robot, the force-reactive policy succeeds in 80% of rollouts, compared with 26.7% for a position-only policy trained on the same demonstrations and executed by the same controller. - [Supporting source](https://arxiv.org/abs/2610.35862) ## Source-stated In simulation, the authors report force tracking with little loss in whole-body motion accuracy. > In simulation, the controller tracks per-hand force commands of varied magnitude and direction, with little loss in whole-body motion accuracy. - [Supporting source](https://arxiv.org/abs/2610.35862) ## Escape Meridian interpretation The reported robot results suggest a possible benefit from using force-reactive control for the tested contact-rich tasks; the abstract does not establish performance on other tasks. > Across three contact-rich tasks on the robot, the force-reactive policy succeeds in 80% of rollouts - [Supporting source](https://arxiv.org/abs/2610.35862) ## Source and contributor context Li, Sihang, Guo, Xiaofeng, Cao, Jinkun, Cheng, Charlie, Yang, Ruihan, Feng, Chen, Shi, Guanya, Qi, Haozhi · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: 1afc0a31-2573-4d6c-8242-793b74d0c721. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-04T23:02:06.691Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 28a78eb94371ce53e1b0064626a93d1a5bf31bb2757e133b2821412a200bc8f2. Source text hash: bcf793aafcec07a74b75babf8f874c0e7b0e8d46b75dc39903c166c0afd417cd. ### uncertain — 2026-11-04T23:02:06.691Z — anthropic/claude-opus-5.5 Publication: 1afc0a31-2573-4d6c-8242-793b74d0c721 attribution — pass: The reading correctly credits FLORE to Sihang Li and coauthors, as listed on the arXiv page. It frames the results as the authors' reports, and the closing inference is labeled as Escape Meridian's reading. > Authors:Sihang Li, Xiaofeng Guo, Jinkun Cao, Charlie Cheng, Ruihan Yang, Chen Feng, Guanya Shi, Haozhi Qi summary — uncertain: The core development and results are faithful, but several central terms are left unexplained for a non-specialist: loco-manipulation, diffusion policy, proprioception, rollouts, and especially measured wrenches, which a lay reader may take to mean tools. The reading also omits the robot-free demonstration method, which used handheld force-sensing grippers and wearable trackers. Smallest repair: briefly gloss these terms or attach tooltips, for example wrenches as forces and torques at the hands, and add one clause on how the demonstrations were collected. > We collect robot-free demonstrations using two force-sensing UMI-style handheld grippers for hand motion and contact wrenches, and wearable trackers for body motion. date — pass: The originalPublished date of 2026-10-26 matches the v1 submission date on arXiv. The later feed entry on 2026-10-27 is correctly not used. > \[Submitted on 26 Oct 2026\] support — pass: Every factual claim matches the abstract, including the 80% versus 26.7% comparison under the same demonstrations and controller, and the simulation force tracking. The interpretation is conditional and labeled, and it does not extend performance beyond the tested tasks. The stated gaps are accurate: the abstract does not name the tasks or report how results vary. > Across three contact-rich tasks on the robot, the force-reactive policy succeeds in 80% of rollouts, compared with 26.7% for a position-only policy trained on the same demonstrations and executed by the same controller. media — pass: The selected media is the verified arXiv abstract page for this paper. It is the direct original source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/15a7d624a7e33081f955475cb7efe403fe80af8a45d71ab624c177bd23a03583/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/sihang-li-and-coauthors-present-flore-for-force-reactive-15a7d624.json)