Research Intern: Self-Aware and Adaptive Multi-Modal Foundation Model-Based Agents - Honda Research Institute USA

Research Intern: Self-Aware and Adaptive Multi-Modal Foundation Model-Based Agents

Your application is being processed

Research Intern: Self-Aware and Adaptive Multi-Modal Foundation Model-Based Agents

Job Number: P26INT-14
Honda Research Institute USA (HRI-US) is seeking a self-motivated research intern to develop and advance adaptive multimodal foundation model-based agents that learn from human demonstrations, interaction, and their own task experience. The intern will investigate test-time adaptation for embodied agents performing procedural tasks. The goal is to enable an agent to form an initial procedure from a human demonstration, revise its behavior using feedback and execution outcomes, and retain useful corrections for future autonomous attempts. The research will explore metacognitive mechanisms that help an agent recognize uncertainty or failure, determine when to seek additional evidence or human feedback, and judge which changes are worth retaining. Internal model representations may provide useful signals alongside observed outcomes. Experiments will primarily use simulated embodied tasks to connect adaptation during interaction with persistent learning across encounters.
San Jose, CA

 

Key Responsibilities

 
  • Develop and evaluate test-time adaptation methods that use demonstration videos, language feedback, and execution history to revise task procedures or action policies during human–agent interaction. 
  • Investigate metacognitive signals for uncertainty, failure detection, and feedback relevance, using model representations and observable task evidence to guide decisions to revise, retry, request feedback, or retain the current behavior. 
  • Explore how useful interaction-conditioned corrections can be selected and consolidated so improvements persist when human feedback is no longer available.
  • Implement reproducible experiments in simulation, comparing adaptation and retention with appropriate baselines and ablations. 
  • Assess task success, feedback and trial efficiency, robustness to imperfect feedback, and retention across task variations; document findings and research contributions.

 

Minimum Qualifications

 
  • Currently enrolled as a Ph.D. student in Computer Science, Machine Learning, Robotics, or a related field. Exceptional M.S. candidates with at least one year of research experience may also be considered.
  • Strong understanding of deep learning and foundation models, with research experience in at least one of multimodal learning, adaptive agents, reinforcement learning, or imitation learning.
  • Proficiency in Python and a deep learning framework such as PyTorch, with experience implementing models and conducting systematic experiments.

 

Bonus Qualifications​

  • Experience with test-time adaptation, in-context learning, meta-learning or meta-RL, policy distillation, or continual learning.
  • Experience with uncertainty estimation, calibration, failure detection, or evaluating the reliability of model-generated predictions and actions.
  • Familiarity with vision–language or vision–language–action models, learning from video demonstrations or human feedback, and embodied simulation environments.
  • Experience probing internal representations or neural activations to understand model behavior and inform adaptation decisions.
  • Strong publication record on topics related to multi-agent systems in robotics or AI conferences (NeurIPS, AAAI, ICLR, AAMAS, RSS, CoRL).

 

Years of Work Experience Required   0
Desired Start Date  May 2027
Internship Duration  3 Months
Position Keywords  Agentic AI, Meta-Cognitive AI

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.