Research Intern: Memory Representation for Dexterous Manipulation

Honda Research Institute USA; San Jose (CA); Sur place; USD 20 000 - 27 000; Plein temps

Résumé du poste

Honda Research Institute USA in San Jose, CA seeks a Research Intern to advance memory representations for dexterous manipulation using multi-modal sensory data. The role focuses on long-horizon, contact-rich tasks and developing latent memories to inform robot policies.

Strong background in representation learning and sequence modeling is required, with experiments in both simulation and real hardware. The position emphasizes publication and patent contribution, with opportunities to publish at

Qualifications

• Ph.D. or MS in robotics, computer science, or a related field.
• Experience with representation learning using multi-modal robot data (e.g., RGB, RGB-D, and tactile).
• Experience with learned state-space models.
• Experience with policy deployment on real-world robots.
• Experience with robotic simulators (e.g., Isaac Gym or MuJoCo).
• Proficient with Python and C++.
• Proficient with PyTorch.
• Experience with Robot Operating System (ROS2).

Responsabilités

• Develop memory representation for long-horizon, contact-rich manipulation for multi-fingered robot hand.
• Design training objectives that enforce information retention over a long-horizon.
• Collect and process data to conduct representation training, and evaluate them on downstream robot policies.
• Implement baselines and perform benchmark evaluations (hardware and simulation).
• Document findings and contribute to internal research reports.
• Contribute to the portfolio of patents, and publish research results when applicable at top-tier conferences and journals in robotics and machine learning.

Connaissances

Representation learning

Multi-modal data

Sequence modeling

Policy deployment on real robots

Python

C++

PyTorch

ROS2

Formation

Ph.D. or MS in robotics, CS, or related field

Outils

Python

C++

PyTorch

ROS2

MuJoCo

Isaac Gym

Description du poste

Research Intern: Memory Representation for Dexterous Manipulation - Honda Research Institute USA

Research Intern: Memory Representation for Dexterous Manipulation

Job Number: P25INT-68

Honda Research Institute USA (HRI-US) is seeking a self-motivated intern to advance memory and belief representations for multi-fingered dexterous manipulation using multi-modal sensory data. The research focuses on long-horizon, contact-rich tasks in which action-relevant information is often only partially observable and must be retained from prior observations. Key objectives include developing learned latent memory representations that preserve information across the timescales required for manipulation and evaluating what information those representations genuinely retain. A successful candidate will have experience in representation learning and sequence modeling, and will evaluate and refine these representations through experiments in simulation and on physical hardware.

San Jose, CA

Key Responsibilities

• Develop memory representation for long-horizon, contact-rich manipulation for multi-fingered robot hand.
• Design training objectives that enforce information retention over a long-horizon.
• Collect and process data to conduct representation training, and evaluate them on downstream robot policies.
• Implement baselines and perform benchmark evaluations (hardware and simulation).
• Document findings and contribute to internal research reports.
• Contribute to the portfolio of patents, and publish research results when applicable at top-tier conferences and journals in robotics and machine learning.

Minimum Qualifications

• Ph.D. or highly qualified M.S. candidate in robotics, computer science, or a related field.
• Experience with representation learning using multi-modal robot data (e.g., RGB, RGB-D, and tactile).
• Experience with learned state-space models.
• Experience with policy deployment on real-world robots.
• Experience with robotic simulators (e.g., Isaac Gym or MuJoCo).
• Proficient with Python and C++.
• Proficient with PyTorch.
• Experience with Robot Operating System (ROS2).

Bonus Qualifications

• Experience with large vision-language-action and world-action models.
• Experience with deep stochastic learning techniques for time series analysis.
• Experience with in-hand dexterous manipulation on multi-fingered robot hands.
• Knowledge in contact dynamics and contact mode switching.
• Experience in online reinforcement learning with hardware robots.
• Experience with Sim2Real approaches.

Years of Work Experience Required 0

Desired Start Date 1/11/2027

Internship Duration 3 Months

Position Keywords Dexterous manipulation, Embodied AI, Robot foundation models