About Flexion

At Flexion, we're building the intelligence layer powering the next generation of humanoid robots. Our mission is to accelerate the transition from fragile prototypes to real-world humanoid deployment. We are founded by leading scientists in robot reinforcement learning (ex-Nvidia, ex-ETH Zürich), and backed by leading international VC firms. In just months, we’ve gone from our first line of code to deploying real humanoid capabilities.

The role

We are looking for a Research Engineer to push the frontier of RL-based manipulation for humanoid robots. You will work on some of the hardest problems in robotics: enabling robots to solve long-horizon, contact-rich manipulation tasks using reinforcement learning, and making these solutions work reliably in the real world. This is a deeply technical role at the intersection of research and engineering.

You will be responsible for developing new algorithms, scaling training systems, and closing the loop between simulation and reality. This includes tackling sparse rewards, exploration, and stability challenges, as well as improving sim-to-real transfer through better modeling, training strategies, and use of real-world data.

You will operate with a high degree of ownership, working closely with simulation, control to build integrated solutions that translate into real robot capabilities.

Key Responsibilities

• Push the frontier of RL-based manipulation. You will design and develop learning-based approaches for complex manipulation tasks, focusing on dexterous, contact-rich interactions. Your work will directly target real-world humanoid capabilities, going beyond benchmark problems toward practical deployment.

• Solve long-horizon and sparse-reward tasks. You will work on algorithms and system designs that enable robots to solve multi-stage manipulation problems with delayed rewards. This includes tackling credit assignment, exploration, and stability challenges in real-world settings.

• Drive sim-to-real transfer. You will develop methods that reliably transfer policies from simulation to physical robots. This includes improving robustness, handling model mismatch, and leveraging domain randomization, system identification, and real-world data.

• Scale reinforcement learning systems. You will extend and optimize large-scale training pipelines, enabling efficient learning across hundreds of thousands of parallel environments. You will work closely with simulation to increase throughput and unlock more complex behaviors.

• Develop new learning approaches. You will explore and implement novel methods at the intersection of reinforcement learning, imitation learning, and model-based techniques, pushing beyond standard approaches when needed.

• Collaborate across the stack. You will work closely with simulation, control to ensure tight integration between learning algorithms and the infrastructure that supports them.