ResearchCareers

Research Fellowship

Explore new robot learning research with enterprise-level compute, alongside a team of ICML authors and NeurIPS award winners.

Apply for the fellowship
Location
Remote
Duration
2–3 months (flexible)
Stipend
$6K / month

Research fellows from

  • ETH Zurich
  • IIT Bombay
  • Yale
  • Stanford
  • Cambridge
01The fellowship

Who should apply

Graduate students and exceptional undergraduates with backgrounds in pretraining large models, reinforcement learning, or adjacent fields, who want to work on general-purpose robotics.

You’ll lead a research project with our team on a challenge we encounter building general-purpose robot models at Pantheon. We will incorporate strong results into the next generation of those models.

02What we offer

Funding and mentorship

Stipend and compute

$6K/month and enterprise-level compute for your experiments.

Weekly 1:1s

Mentorship from a team including ETH Zurich alumni, ICML authors, NeurIPS award winners, and researchers from xAI and Jane Street.

Public research

Work toward a paper or other public research output.

Full-time roles

Outstanding fellows will be offered full-time roles at Pantheon after the fellowship.

03Research directions

Current research questions

These questions come directly from training and evaluating our general-purpose robot models: one policy that has to handle many tasks, objects, and stations rather than a single fixed one.

  1. 01

    Value learning

    How can we learn a value function for general purpose tasks?

  2. 02

    Model drift & test-time adaptation

    How should world models detect and adapt when predictions diverge from experience?

  3. 03

    Multimodal representation

    How should world models combine vision, touch, force, and proprioception when signals arrive at different rates or are missing?

  4. 04

    Scaling laws for Forward Dynamics Models

    How do data, compute, and model size trade off in robot learning, and how does this compare with video pretraining?

  5. 05

    Learning from failure

    How do we deliberately induce and learn from failures, and does failure data beat an equal budget of successful demonstrations?

04How to apply

Apply to the fellowship

Send us your CV, a short note on the research questions that excite you, and any relevant work to fellowship@pantheon.inc.

Apply for the fellowship
© 2026 Pantheon Industries, Inc.