Senior Reinforcement Learning Engineer

Gravis Robotics is a high-growth Series A start-up backed by SoftBank, bringing Physical AI to the construction industry, turning heavy construction machines into autonomous robots. Gravis began as an ETH Zurich spin-out, and our unique combination of learning-based automation and augmented remote control lets one operator safely conduct a fleet of earthmoving machines in a gamified environment. Backed by deep robotics research and now deployed across multiple countries with leading construction and equipment partners, our team is rapidly growing to bring this technology to a trillion-dollar industry. The Gravis RACK is a machine-agnostic retrofit kit that adds autonomy to excavators and wheel loaders from 10 to 100+ tonnes: LiDAR and camera sensing, GNSS RTK, networking hardware and rugged edge compute that works offline. Paired with the Slate tablet and our Copilot software, it lets an operator run a machine manually, with AI assistance, or fully autonomously. Increasingly, we also build custom hardware to adapt our machines for highly specialized, robust applications beyond traditional excavation. About the Job The autonomy team at Gravis builds autonomous systems for excavators operating in real construction environments. You will build control modules that run on many different machines , across many sites, with different soil conditions. We’re looking for a roboticist with data driven planning and/or control background, deep python expertise and good level of C++ proficiency. To be successful in this role you should have experience working with real robots, tackling the challenges of sim2real transfer, and deploying robotic systems in a production environment. What you will do Learning-Based Planning and Control for Real Systems Develop data driven planning and control systems for autonomous excavation that generalize across machine models and soil conditions Contribute to simulation improvements that reduce or address the sim2real gap Define data collection and curation pipelines for incorporating real data in policy training Design experiments focused on continuous performance and robustness improvements. Explore the usage of adaptive and online reinforcement learning in deployed systems Provide mentorship and supervision for junior team members, interns, and students. System Integration Integrate learned components into a larger software stack Collaborate with excavation and motion planning engineers Build tools for analysing and evaluating the behavior of learned components What we’re looking for We recognize that excellent candidates come from diverse backgrounds with various combinations of skills. If you meet most of the core qualifications below, we highly encourage you to apply. Core qualifications 2–5 years industry experience developing Reinforcement learning systems for control and/or planning and deploying them on real robots with a customer. If you only have experience with simulation, you’re most likely not a good fit for this position. Experience with GPU accelerated simulation environments (e.g. IsaacSim/IsaacLab, CARLA, MuJoCo) Strong Python skills and experience with PyTorch or similar libraries Proficiency in C++ Comfortable debugging real-world system behavior Ability and willingness to travel as required by business projects. Great-to-Have Skills & Experience Experience with hydraulic machinery Experience with supervised learning or imitation learning Research experience in reinforcement learning Experience deploying robotic systems at scale (e.g. hundreds of units) Familiarity with ROS or similar robotics frameworks Experience with feature-flagged deployments, staged rollouts, or long-lived platforms Experience with data curation for ML applications Experience guiding, mentoring, or leading junior colleagues, students,