Robocurve GTM brain

Robocurve develops independent real-world robotics evaluations and open-source tools for AI labs and robotics teams.

category

Physical AI evaluation

market profile

  • text: Real-robot tasks expose capability and safety performance that simulated tests or curated demonstrations may miss.
  • text: Robotics models can be compared on task success, cost per attempt and time, making evaluation a model-selection input.

personas

  • title: Head of robotics research
    pains:
    • Needs repeatable physical tests to compare checkpoints and identify failure modes
    • Maintaining robots and benchmark environments takes specialist time
  • title: AI safety and evaluations lead
    pains:
    • Cannot infer physical-world safety from language-only evaluations
    • Needs evidence on refusal and unsafe instruction handling
  • title: Robot product or procurement lead
    pains:
    • Vendor demos do not provide comparable evidence of actual task performance

icp

  • text: Frontier AI labs and robotics model developers with physical policies to test, regardless of company size.
  • text: Academic robotics groups building reproducible physical benchmarks are prospective open-source users and partners.

message pillars

  • text: Measure capabilities in physical settings.
  • text: Assess how models respond to harmful requests.
  • text: Publish evidence for the broader public.

objections

  • objection: Will benchmark tasks reflect our deployment environment?
    response: Review task definitions, hardware and failure modes, then scope a representative pilot before generalizing the scores.
  • objection: Why use an independent evaluator when we already have internal tests?
    response: Independent repeatable tests add a common comparison point; use them alongside internal deployment-specific validation.

competitors

  • name: Positronic Robotics
    why: Provides independent physical AI benchmarking on real hardware.
  • name: Coop Labs
    why: Runs physical robot benchmarks for policy comparison and failure analysis.
  • name: Enact
    why: Builds physical environments for robotics model benchmarking.

keywords

  • robotics evaluation
  • physical AI benchmark
  • robot policy evaluation
  • VLA benchmark
  • robot safety testing
  • real world robotics testing

voice

  • empirical
  • public-interest oriented
  • technical

signals

  • text: A robotics lab is releasing new policy checkpoints and needs comparable independent results.
  • text: A research group is building a physical benchmark or looking for open-source evaluation infrastructure.

stack

  • Inspect Robots

hiring

  • Member of Technical Staff, Frontier Robotics Evals
  • Member of Technical Staff, Frontier Robotics Evals (Biosecurity)

pricing

model: unknown

motion

hybrid

channels

  • open source
  • research publications
  • X
  • LinkedIn

confidence

medium

Founders

  • name: Jay Chooi
    title: Founder/CEO
    role: CEO
    bio: Founder and CEO of Robocurve. MA Statistics and BA CS/Math from Harvard. Jay builds evals of frontier AI and forecasts its progress. Previously Research Fellow at MATS, researcher at the UK AI Security Institute, and top contributor to Inspect Evals, the UK government's AI eval framework. Published at ACM EC, ICML, ACL, and EMNLP. Called the 2024 election correctly in all 50 states. Won a Rhodes Scholarship. Gold medal at the International Olympiad on Astronomy and Astrophysics.

Back to YC companies