Research Intern: Closed-loop E2E Driving

Honda Research Institute USA; Mountain View (CA); Sur place; USD 35 000 - 55 000; Plein temps

Résumé du poste

Honda Research Institute USA in Mountain View, CA is seeking a PhD-level research intern to advance closed-loop E2E driving, spanning policy learning, trajectory generation, and training/evaluation in a state-of-the-art simulation bench.

The internship offers hands-on experience with cutting-edge autonomous driving research, multi-GPU training, and collaboration with full-time researchers, with opportunities for publications and patents.

Qualifications

• Currently enrolled in a PhD program in Robotics, Computer Science, Electrical Engineering, or a related field.
• Solid foundation in deep learning and strong proficiency in Python and PyTorch.
• Understanding of motion planning and/or policy learning fundamentals (imitation learning, reinforcement learning, or trajectory optimization).
• Experience training and debugging neural networks at scale.
• Ability to read, implement, and extend recent research papers independently.

Responsabilités

• Research and develop planning modules for E2E driving models, including trajectory scoring/selection and multi-modal planning heads.
• Design closed-loop training strategies to mitigate distribution shift and errors.
• Build and run closed-loop evaluation pipelines on benchmarks and analyze failures.
• Conduct ablations and large-scale experiments; optimize training/inference throughput on multi-GPU clusters.
• Collaborate with researchers and engineers; present findings internally and contribute to publications or patents.

Connaissances

Deep learning

Python

PyTorch

Motion planning

Formation

PhD student in Robotics/CS/EE

Outils

CUDA/TensorRT

Multi-GPU training

Description du poste

Research Intern: Closed-loop E2E Driving - Honda Research Institute USA

Research Intern: Closed-loop E2E Driving

Job Number: P25INT-57

Honda Research Institute USA (HRI-US) is seeking a talented and motivated PhD-level research intern to explore methods to improve closed-loop planning performance — spanning policy learning, trajectory generation, and closed-loop training/evaluation — and validate them in state-of-the-­art simulation benchmark. This internship is ideal for PhD students with a strong publication record in deep learning and planning who are looking to apply their research to practical robotics and autonomous systems.

Mountain View, CA

Key Responsibilities

• Research and develop planning modules for E2E driving models, including trajectory scoring/selection, diffusion- or anchor-based trajectory generation, and multi-modal planning heads
• Design closed-loop training strategies (e.g., closed-loop fine-tuning, RL/IL hybrid post-training, world-model-based rollout training) to mitigate distribution shift and compounding errors
• Build and run closed-loop evaluation pipelines on benchmarks such as NAVSIM, Bench2Drive, CARLA Leaderboard, or nuPlan; analyze failure modes and drive iterative improvement
• Conduct ablations and large-scale experiments; profile and optimize training/inference throughput on multi-GPU clusters
• Collaborate with full-time researchers and engineers; present findings internally and contribute to publications and/or patents

Minimum Qualifications

• Currently enrolled in a PhD program in Robotics, Computer Science, Electrical Engineering, or a related field.
• Solid foundation in deep learning and strong proficiency in Python and PyTorch.
• Understanding of motion planning and/or policy learning fundamentals (imitation learning, reinforcement learning, or trajectory optimization).
• Experience training and debugging neural networks at scale.
• Ability to read, implement, and extend recent research papers independently.

Bonus Qualifications

• Publications at top-tier venues (CVPR, ICCV, ECCV, NeurIPS, ICML, ICLR, CoRL, ICRA, RSS)
• Hands‑on experience with E2E driving frameworks (e.g., UniAD, VAD, PARA-Drive, DiffusionDrive, Hydra-MDP) or closed-loop benchmarks (NAVSIM, Bench2Drive, CARLA, nuPlan)
• Experience with generative planners (diffusion policies, flow matching), world models, or RL post‑training of driving policies
• Familiarity with simulation infrastructure, sensor simulation, or scenario generation
• Track record in autonomous driving challenges/leaderboards (e.g., CARLA Leaderboard, Waymo/nuPlan challenges)
• Strong software engineering practices (CUDA/TensorRT optimization, distributed training, CI)

Years of Work Experience Required 0

Desired Start Date 1/25/2027

Internship Duration 3 Months

Position Keywords E2E, Closed loop simulation, Autonomous Driving