Reinforced Learning Engineer

About the Role We build autonomous robotic systems spanning legged robots and underwater/surface vehicles (AUVs/USVs). We are seeking a Senior RL Engineer to design, train, and deploy reinforcement learning policies that run on real hardware. This is a hands-on role in an early-stage, hardware-in-the-loop environment where policy development and mechanical bring-up happen in parallel. You'll own the full pipeline from environment design and training through sim-to-real transfer, embedded deployment, and field validation.

Key Responsibilities

• Design, train, and tune RL policies for locomotion (quadruped) and vehicle control (AUV/USV)
• Build and customize simulation environments and reward formulations (Isaac Lab / Isaac Sim, MarineGym, or equivalent)
• Develop robust sim-to-real transfer pipelines — domain randomization, system identification, and observation/action-space design matched to real hardware
• Export and deploy trained policies to embedded compute (e.g., Jetson) and validate on real robots
• Design observation spaces consistent with real state estimation — IMU/DVL/depth fusion, accounting for sensor noise and estimator latency
• Implement safe RL practices — constrained policy optimization, fault detection, fallback/recovery behaviors, and policy-level integration with watchdog and E-stop systems
• Handle partial observability and disturbance rejection, particularly for underwater operation (currents, hydrodynamic disturbances, sensor dropout)
• Collaborate with mechanical, embedded, and controls engineers to align training configs with real actuator, sensor, and dynamics behavior
• Profile and optimize inference for real-time control loops
• Establish evaluation, logging, benchmarking, and experiment-tracking practices; maintain reproducible, versioned training pipelines
• Document training pipelines, configs, and deployment procedures

Required Qualifications

• Bachelor's or Master's degree in Robotics, Computer Science, Machine Learning, Electrical/Mechanical Engineering, or a related field
• Minimum 4 years of experience in reinforcement learning and/or robot learning
• Demonstrated sim-to-real deployment of RL policies on physical hardware (required)
• Strong proficiency in Python and a deep RL framework (PyTorch; RL libraries such as RSL-RL, Stable-Baselines3, or equivalent)
• Hands-on experience with robotics simulators (Isaac Lab/Isaac Sim, MuJoCo, or similar)
• Solid understanding of robot dynamics, control, and the practical challenges of the reality gap
• Experience deploying models to edge/embedded compute for real-time inference
• Comfort debugging across the sim / embedded / hardware boundary

Preferred Qualifications

• Experience with legged locomotion RL and/or underwater vehicle control
• Familiarity with underwater RL environments (MarineGym) and marine autonomy stacks (ROS 2, MOOS-IvP, DUNE/Neptus)
• Experience with domain randomization, privileged learning, or teacher-student architectures
• Knowledge of curriculum learning and multi-task or multi-embodiment transfer
• Background in motor control, CAN-based actuator interfaces, or real-time systems
• Experience with sensor fusion and state estimation (IMU/DVL/depth)
• Familiarity with MLOps tooling — experiment tracking (Weights & Biases, TensorBoard), distributed training, and CI for policy regression testing
• Publications or open-source contributions in robot learning resumes to mohan.krishnan@xerarobotics.com