Postdoctoral Scientist: Multi-Modal Representation Learning for Dexterous Manipulation - Honda Research Institute USA

Postdoctoral Scientist: Multi-Modal Representation Learning for Dexterous Manipulation

Your application is being processed

Postdoctoral Scientist: Multi-Modal Representation Learning for Dexterous Manipulation

Job Number: P24T06
Honda Research Institute USA (HRI-US) in San Jose, California, is seeking a highly motivated postdoctoral scientist to join our robotics research team. This role focuses on advancing dexterous robotic manipulation through the development of multi-modal representations that integrate tactile, force, vision, audio, and language inputs. We are particularly interested in candidates who can leverage and adapt large models—such as vision-language models, multi-modal transformers, or foundation models—to build rich representations of robot-object interactions and integrate them into action policies that demonstrate their effectiveness in multi-fingered dexterous manipulation. The successful candidate will contribute to algorithm development in both simulation and real hardware, integration with advanced robotic hands, and the collection of multi-modal datasets for manipulation. Ideal candidates will have a strong background in tactile sensing, ROS, and multi-sensory machine learning algorithms for robotic manipulation and control.
San Jose, CA

 

Key Responsibilities

 
  • Design and develop multi-modal representation learning algorithms that integrate tactile, force, vision, audio, and language inputs for contact-rich robot-object interaction.
  • Adapt and fine-tune large models (e.g., vision-language models, multi-modal transformers) for robotic perception and action in dexterous manipulation scenarios.
  • Develop action policies or control heads that directly leverage learned representations to perform precise, in-hand, and contact-rich manipulation using multi-fingered hands.
  • Deploy perception and control algorithms on real robotic platforms, including sensor calibration, system integration, and testing on state-of-the-art multi-fingered hands.
  • Collect and curate multi-modal datasets from both physical hardware and simulation environments to support model training, benchmarking, and sim-to-real transfer.
  • Publish research results in top-tier conferences and journals in robotics and machine learning.

 

Minimum Qualifications

 
  • Ph.D. in computer science, robotics, or a related field.
  • Experience working with tactile, force, and/or visual sensing in robotic manipulation tasks.
  • Strong background in machine learning and experience with deep learning frameworks such as PyTorch.
  • Experience in implementing, training, and deploying machine learning models on physical robotic hardware
  • Strong programming skills in Python or C++.
  • Familiarity with ROS and developing end-to-end systems.

 

Bonus Qualifications

  • Experience with robotic manipulation, grasping, and control using multi-fingered hands.
  • Hands-on experience with tactile sensing technologies (e.g., taxel-based sensors, GelSight).
  • Familiarity with simulation platforms such as Isaac Sim or Isaac Lab.
  • Experience building digital twins or sim-to-real pipelines for robotics.
  • Experience with computer vision and multisensory perception.
  • Experience with large models (e.g., VLMs, multi-modal transformers, or foundation models) for perception or control.

 

Desired Start Date 11/3/2025
Contract Duration 3 years
Position Keywords Robotics, Perception, Object Manipulation, Representation Learning, Vision, Tactile, VisuoTactile, Vision-Language Models, Deep Learning

Alternate Way to Apply

Send an e-mail to careers@honda-ri.com with the following:
- Subject line including the job number(s) you are applying for 
- Recent CV 
- A cover letter highlighting relevant background (Optional)

Please, do not contact our office to inquiry about your application status.