Relari

Relari's core claim is a data argument, not a model argument, and that is what makes it interesting. The company's own framing is that video scales best but carries the weakest interaction signal, teleoperation carries the richest signal at the lowest scale, and video plus a tactile glove is rich but hard to scale. EMG plus video is meant to sit in the middle by moving force supervision off instrumented objects, gloves, and lab hardware and onto the demonstrator's arm. The concrete signals it decodes are specific: grip force distribution across fingers (from EMG patterns of how effort is spread), stiffness (from muscle co-contraction as the hand braces or yields), anticipation (muscle activity before contact and the force being prepared), and reflex (rapid response to slips and mistakes). A hand can look still in video while muscles continuously regulate force, contact, and stability — that is the gap Relari targets. The stated claim is that once decoded, this biomechanical structure transfers across robot embodiments into force-aware control. That is the important word: asserted, not demonstrated in the material available. No accuracy figures, no cross-subject generalization results, and no benchmarks against teleoperation are published, and cross-embodiment transfer is the hardest part of the pipeline. EMG itself is also notoriously sensitive to sensor placement, skin condition, and the individual demonstrator, and none of that variability is quantified publicly. Where it fits: robotics labs studying contact-rich manipulation, teams building general-purpose manipulation foundation models, biomechanics groups that already own EMG instrumentation and know its failure modes, and investors tracking physical AI data pipelines. Where it does not: robotics teams that need a commercially available system now, developers who want plug-and-play control endpoints, and any buyer running a procurement process that requires published commercial terms. If you have the research bench to co-develop, the collaboration is credible; if you need to ship a manipulation capability this quarter, Relari is not that vendor. Backing from Y Combinator, Soma Capital, and General Catalyst at least signals the research direction has institutional validation.