Simulating human-robot decision making for embodied AI

This article is a part of Startup Spotlight, a series that features young, up-and-coming startups.

Image credit: Timmy Loen

During a meeting with AI pioneer Fei-Fei Li, Nicolas Duval asked why crowdsourced teleoperation projects avoided software environments. She explained that simulating perfect physical contact remained nearly impossible.

Duval realized that human-robot decision-making could scale digitally even if physical dexterity could not. He launched Synthium to capture this critical interaction data at scale.

😟 Problem

Embodied AI systems require massive datasets to learn complex behaviors. Traditional real-world teleoperation costs over US$100 per usable hour.

This high cost severely bottlenecks the development of humanoid robots. Hardware-based data collection methods remain too slow to meet surging industry demands.

💡 Solution

Synthium builds a human-in-the-loop software platform to train autonomous AI models. The system generates structured datasets before deploying robots in physical environments.

The platform extracts this training data through three core mechanisms:

• Runs continuous simulation sessions where live humans interact with robotic avatars.
• Captures multimodal data across motion, voice, language, and human reasoning.
• Structures raw interactions into clean training sets for Vision-Language-Action models.

Image credit: Synthium

📊 Market size

The current serviceable data market for embodied AI companies sits between US$0.9 billion and US$1.2 billion. Around 240 active programs maintain annual data budgets up to US$5 million.

This specific data segment is projected to reach US$20 billion by 2034. Top-down estimates project the broader humanoid robot market will reach US$38 billion by 2035.

🤝 Team

• Nicolas Duval. Co-founder and CEO. He previously worked as a software architect at Autodesk and Adobe.
• Nicolas Savva. Co-founder and CTO. He holds a PhD from Cornell University and specializes in computer graphics and computer vision infrastructure.

🚀 Traction

🏆 Competition

💰 Financials

🚩 Risks