One Robot GTM brain

category

market profile

• text: Robotics teams face limited physical robot time and costly reset cycles when evaluating action policies.
• text: World-model simulation is an emerging alternative to physical evaluation, but the connection between simulated and real-world performance must be validated for each use case.
• text: Contact-rich manipulation and deformable materials pose challenges for simulation of realistic interactions.

personas

• title: Head of robotics or autonomy
  pains:

  • Access to physical robots and lab time limits training iterations
  • Policy failures need to be found before costly deployment
• title: Robot learning research lead
  pains:

  • Collecting diverse real-world trajectories is expensive
  • Simulated manipulation may fail to capture contact and deformation
• title: Manipulation engineering lead
  pains:

  • Reproducing rare failures on physical hardware takes repeated setup and resets
  • Training data must target the failure modes that matter for a specific task

icp

• text: Robotics developers building action models for manipulators in material handling, assembly or deformable-object tasks.
• text: Teams with robot experience data and policies to evaluate that need more experiments than their physical lab capacity permits.
• text: Industrial robotics companies moving learned manipulation toward deployment and needing evidence that simulated evaluations correlate with hardware.

message pillars

• text: Learn a task-specific simulation from robot data.
• text: Use simulated rollouts to identify weaknesses.
• text: Produce synthetic trajectories that target observed failures.
• text: Focus on difficult physical interactions.

objections

• objection: Will simulation transfer to our robot and task?
  response: Run a representative pilot and compare simulation results with held-out physical trials before relying on policy rankings.
• objection: How much data is required to build our environment?
  response: Discuss available robot observations and actions with the team; public material does not state a minimum dataset.
• objection: Does synthetic data actually improve deployment performance?
  response: Measure improvement against a physical baseline for the same task, including failure rates and test conditions.

competitors

• name: Runway
  website: https://runway.com/research/accelerating-robot-policy-evaluation
  why: Researches world-model-based robot policy evaluation as an alternative to hardware testing.
• name: NVIDIA
  why: Isaac Sim and related robotics simulation tooling support policy training and evaluation.
• name: DexForce
  website: https://www.reuters.com/technology/chinese-startup-dexforce-trains-humanoid-robots-using-virtual-data-slash-costs-2026-10-15/
  why: Builds virtual training environments for robotics, including industrial tasks.

keywords

• robot world model
• robot manipulation simulation
• robot policy evaluation
• synthetic robot training data
• deformable object simulation
• sim to real robotics
• robot learning edge cases

voice

• Technical and concise
• Experiment-oriented

signals

• text: A robotics team is collecting manipulation data and reports physical robot availability as its training bottleneck.
• text: A team is testing learned policies for contact-rich assembly or deformable objects.

pricing

motion

channels

• founder email
• YC directory

confidence

Founders

• name: Hemanth Sarabu
  title: Founder/CEO
  role: CEO
  bio: Bringing robots to life using world models and machine learning. Previously built robot learning and perception systems at Industrial Next, Symbio Robotics, NASA JPL, and Google. Bootstrapped geospatial AI company, Crescer AI, to profitability.
• name: Elton Shon
  title: Founder/CTO
  role: CTO
  bio: Robotics, SW/FW. Previously built robot learning and control system at Industrial Next and helped build Dojo at Tesla.