Research Intern: Life-Long Embodied Spatial Reasoning

Honda Research Institute USA; San Jose (CA); On-site; USD 34,000 - 55,000; Part time

Job summary

Honda Research Institute USA in San Jose, CA is offering a Research Intern position focused on Life-Long Embodied Spatial Reasoning. The internship targets building capable agents that continually build, update, and reason over spatial knowledge across episodes, integrating memory for planning and action in dynamic environments.

The role spans exploration of embodied navigation, 3D scene representation, continual learning, long-term memory systems, and multimodal foundation models, with hands-on

Qualifications

• PhD or near completion in CS/AI/Robotics.
• Extensive research background in CS, deep learning and 3D computer vision.
• Strong programming skills in Python with ML libraries (PyTorch, TensorFlow).

Responsibilities

• Conduct research on spatial reasoning and long-context video understanding.
• Review and reproduce relevant state-of-the-art research.
• Design and implement novel models, representations or memory architectures.
• Develop evaluation tasks and benchmarks for long-horizon spatial reasoning.
• Train and evaluate models using large-scale video and multimodal datasets.
• Analyze model behavior and identify limitations in spatial and temporal reasoning.
• Collaborate with researchers and engineers to prototype and validate research ideas.
• Communicate research findings through presentations, technical reports, and publications.

Skills

Python

Deep learning

3D computer vision

Education

Ph.D. in CS/AI/Robotics

Tools

PyTorch

TensorFlow

Job description

Research Intern: Life-Long Embodied Spatial Reasoning - Honda Research Institute USA

Research Intern: Life-Long Embodied Spatial Reasoning

Job Number: P26INT-11

Honda Research Institute USA (HRI-US) is seeking a research intern in lifelong embodied spatial reasoning to explore new approaches for enabling embodied agents to continually build, update, and reason over spatial knowledge of their environments through sustained interaction. Modern vision-language and embodied foundation models can navigate and answer spatial questions within a single episode. However, agents operating in the same environments over days and weeks require additional capabilities, such as consolidating experiences across episodes into compact, persistent spatial representations; detecting and incorporating changes as objects move and scenes evolve; actively exploring to resolve uncertainty and fill gaps in their knowledge; and grounding language instructions and queries in accumulated memory to plan and act. The intern will investigate methods at the intersection of embodied navigation, 3D scene representation, continual learning, long-term memory systems, and multimodal foundation models.

San Jose, CA

Key Responsibilities

• Conduct research on spatial reasoning and long-context video understanding.
• Review and reproduce relevant state-of-the-art research.
• Design and implement novel models, representations, or memory architectures.
• Develop evaluation tasks and benchmarks for long-horizon spatial reasoning.
• Train and evaluate models using large-scale video and multimodal datasets.
• Analyze model behavior and identify limitations in spatial and temporal reasoning.
• Collaborate with researchers and engineers to prototype and validate research ideas.
• Communicate research findings through presentations, technical reports, and potentially research publications.

Minimum Qualifications

• Currently pursuing or recently completed a Ph.D. degree in Computer Science, Artificial Intelligence, Robotics, or a related field.
• Extensive research background in computer science, deep learning, and 3D computer vision.
• Strong programming skills in Python, with experience in machine learning libraries such as PyTorch, TensorFlow or similar libraries.

Bonus Qualifications

• N/A

Years of Work Experience Required 0

Desired Start Date May 2027

Internship Duration 3 Months

Position Keywords Embodied AI, Spatial AI, Spatial Reasoning, Long Video Understanding