A Zurich spin-off founded in 2024, Flexion Robotics is working to resolve the main obstacle in humanoid robotics, namely the absence of a layer of intelligence capable of adapting a robot to varied tasks without requiring human supervision. The startup designs a reinforcement learning and simulation platform intended to produce generalizable autonomous behaviors, breaking with the rigid and scripted architectures still dominant in the industry.
Its approach is based on a technological stack structured in three layers. The first, dedicated to control, uses language models to interpret instructions in natural language, divide them into subtasks and anchor them in the real environment. The second, devoted to movement, is based on a vision-language-action model trained mainly on synthetic data before being refined on the limit cases of the physical world. The third, finally, manages low-latency full body control via a library of modular skills based on transformers, allowing new behaviors to be quickly composed.
Flexion Robotics wants to respond to the structural labor shortage and demographic aging which are accelerating the demand for robots capable of performing complex manual tasks. The startup is already collaborating with several leading equipment manufacturers to test and deploy its models in industrial conditions.
This work builds on Nikita Rudin and David Hoeller, who led the development of Isaac Lab at NVIDIA, which has become a global benchmark for reinforcement learning in simulation. Their simulation-first approach, generative models and real-time controllers aim to accelerate learning, reduce data collection costs and improve transferability from virtual to real.
In this context of the rise of humanoids and the convergence between generative artificial intelligence and robotics, Flexion Robotics intends to position itself as the supplier of the central intelligence layer, a strategic role in a sector undergoing industrialization.
The startup has just raised $50 million in Series A with DST Global Partners, NVentures (NVIDIA), Prosus Ventures, Redalpine and Moonfirea few months after a first round of 7.35 million. This funding should make it possible to strengthen R&D in Zurich, expand computing capacities and the robotic fleet, establish a presence in the United States and accelerate the commercialization of its autonomy stack.
