Principal Architect - Reinforcement Learning (Physical AI)
Core
Design and deploy end-to-end reinforcement learning systems that train perception, prediction, planning, and control policies for heavy machinery (excavators, dozers, loaders, haul trucks) using simulation and fleet data.
Role type
Principal Architect, Reinforcement Learning (Physical AI)
Builds
High-performance RL and imitation learning pipelines for autonomous construction equipment
Domain
Heavy machinery / Construction autonomy / Physical AI
Deliverable
production ML models
Required skills
Reinforcement Learning (PPO, SAC, GRPO, model-based/world-model RL), Imitation Learning (behavior cloning, offline RL, distillation), Large-scale closed-loop training in simulation (GPU-parallel, Ray, Kubernetes), Sim-to-real transfer strategies, Policy evaluation and safety metrics, Controls and Robotics foundations (MPC, trajectory optimization)
Preferred skills
Hydraulic or heavy-machinery dynamics expertise, Foundation-model post-training (RLHF, DPO)
Technologies
NVIDIA Isaac Lab, Isaac Sim, Omniverse, NVIDIA Cosmos, MuJoCo, Brax, Ray, Kubernetes, Slurm
Responsibilities
Define reference architecture for learning-based autonomy; design task formulations, reward structures, and curricula; architect massively parallel GPU-based RL training; establish sim-to-real transfer methodologies; develop policy evaluation and safety-case frameworks; partner on large-scale training and evaluation operations
Seniority
Principal, hands-on IC with strategy & mentorship
