The data problem in robotics is a bottleneck measured in years. Deploying a robot for a new task can take months of manual programming and thousands of hours of trial-and-error data collection. Deplace AI, a Paris-based startup founded in 2025, is betting that the solution is already being recorded on millions of smartphones. Its core proposition: turn ordinary human motion videos into the structured language needed to train and evaluate so-called Physical AI [F6S, 2025].
The Wedge: From Video to Vocabulary
Deplace AI's initial product surface is a platform that benchmarks multiple AI robots on real-world workflows, claiming to help companies select the right hardware [Deplace AI, 2025]. The deeper technical wedge, however, is the data pipeline underneath. The company is building a Motion2Text API designed to ingest raw human video and output detailed, language-level descriptions of the actions performed [LinkedIn, 2026]. This translation from pixels to prose is meant to create a scalable, high-quality training dataset for robotic control models, addressing a critical scarcity in the field.
The Team and Technical Credentials
The founding trio brings a mix of technical and entrepreneurial focus. CEO Pedro Milcent leads the company, having described its mission as "bringing robots to life with human data" [F6S, 2025]. Co-founder Elie Chelly is a PhD candidate in robotics, tactile sensing, and dexterous manipulation at ISIR Sorbonne, providing deep academic grounding in the problem space [LinkedIn, 2026]. Both Chelly and co-founder Nathan Kammoun studied at EPFL, a Swiss university with a strong robotics pedigree [LinkedIn, 2026]. The team's public profile is early-stage, but the technical composition aligns with the deeptech challenge.
| Founder | Role | Key Background |
|---|---|---|
| Pedro Milcent | Co-founder, CEO | Mission focus on human data for robotics [F6S, 2025]. |
| Elie Chelly | Co-founder | PhD candidate in robotics & dexterous manipulation at ISIR Sorbonne [LinkedIn, 2026]. |
| Nathan Kammoun | Co-founder | Studied at EPFL [LinkedIn, 2026]. |
Early Traction and Claims
Deplace AI's public traction is anchored in a single, named case study and a set of self-reported performance metrics. The company states it achieved measurable throughput gains for a leading e-commerce fulfillment company by identifying and deploying the right AI robot, doing so with "zero operational disruption" [Deplace AI, 2025]. Its broader platform claims are ambitious:
- Performance lift. Robots evaluated on actual workflows see 40% higher performance, according to the company [Deplace AI, 2025].
- Deployment speed. The data-backed selection process leads to a 6x faster time-to-deployment [Deplace AI, 2025].
- Infrastructure burden. The platform handles the entire evaluation setup with zero operational burden on the client [Deplace AI, 2025]. These figures, while unverified by third parties, outline the value proposition: faster, better-fitting robotic deployments.
The Competitive and Technical Hurdles
The space for robot simulation and evaluation is not empty. Established players like inVia Robotics offer integrated robotics-as-a-service for fulfillment. Startups like Antioch operate in the physical AI simulation arena. Deplace AI's differentiation rests on its proprietary data translation layer,the Motion2Text API,rather than on building robots themselves. The primary risk is technical validation at scale. Converting diverse, unstructured human video into precise, actionable language for robots is a formidable machine learning challenge. A secondary risk is commercial: the platform's utility depends on attracting both robot makers and end-users to its benchmarking ecosystem before either group commits to a single hardware vendor's own tools.
The Next Twelve Months
With a disclosed $200,000 pre-seed from the Berkeley SkyDeck Fund, the company is in the earliest stage of proving its thesis [Berkeley SkyDeck, Unknown]. The immediate milestones are clear. First, move beyond the initial e-commerce case study to publicly name additional pilot customers. Second, release technical details or a limited API for the Motion2Text technology to demonstrate its robustness to the developer community. Third, secure a seed round to scale the data collection and engineering effort required to move from a promising wedge to a broad platform. The bet from SkyDeck is a start, but the next check will need to come from investors convinced that human video is the missing dataset that finally unlocks scalable Physical AI. Can a pipeline built in Paris become the lexicon for the world's robots?
Sources
- [Berkeley SkyDeck, Unknown] Berkeley SkyDeck Fund | https://skydeck.berkeley.edu/
- [Crunchbase, Unknown] Deplace AI - Crunchbase Company Profile & Funding | https://www.crunchbase.com/organization/deplace-ai
- [Deplace AI, 2025] Deplace AI homepage | https://deplaceai.com/
- [F6S, 2025] Deplace AI company profile | https://www.f6s.com/company/deplace-ai
- [LinkedIn, 2026] Deplace AI company page | https://www.linkedin.com/company/deplace-ai