Relari GTM brain

Relari researches robot manipulation models trained on human demonstrations and biomechanical signals for potential robotics partners.

category

Robot learning and dexterous manipulation

market profile

  • text: Robot foundation model developers need diverse physical interaction data, while force and contact are difficult to infer from video alone.
  • text: Robot training data is a bottleneck and companies pursue different combinations of human video, teleoperation and wearables.
  • text: Relari presents a research direction rather than a publicly described commercial offer or customer list.

personas

  • title: Head of robot learning
    pains:
    • Collecting force-rich demonstrations across many tasks can be slow and tied to specific robots.
    • Vision-only datasets miss grip force and contact information.
  • title: Robotics research director
    pains:
    • A learned manipulation policy must transfer to new robot hands and tasks.
    • Data collection needs to scale beyond one instrumented laboratory.
  • title: Automation engineering leader
    pains:
    • Variable manipulation requires more adaptable control than a fixed routine.
    • A new model needs validation on the actual robot and target workflow.

icp

  • text: Hypothesis: robotics developers building dexterous hands or grippers and investing in learned manipulation; exact customer profile is not published.
  • text: Hypothesis: physical AI labs that need multimodal human demonstration data across varied tasks and environments.

message pillars

  • text: Reveal physical effort missing from video.
  • text: Capture interaction data with a wearable rather than instrumenting every object.
  • text: Train force-aware manipulation models from human dexterity.

objections

  • objection: Will the learned policy work on our robot?
    response: Treat transfer as a research hypothesis and evaluate success rates on the intended embodiment and tasks.
  • objection: Is a wearable worth adding to video capture?
    response: EMG can add muscle activity and force timing that video does not directly reveal; measure the improvement against the collection cost.
  • objection: Is there a product we can purchase now?
    response: The site describes research and hiring, not a public product or commercial terms; ask the company about availability.

competitors

  • name: Physical Intelligence
    why: Develops general robot foundation models for manipulation across hardware.
  • name: Skild AI
    why: Develops a robot foundation model for multiple robot embodiments.
  • name: Dyna Robotics
    why: Trains robot manipulation models using human video and other data.

keywords

  • robot dexterous manipulation
  • robot foundation models
  • EMG robot learning
  • force aware robot control
  • human demonstration robotics
  • robot manipulation training data
  • cross embodiment robot learning

voice

  • research-led
  • technical
  • exploratory

signals

  • text: A robotics lab is hiring for manipulation learning, dexterous hands or human demonstration datasets.
  • text: A robot maker is seeking to scale contact-rich training beyond teleoperation in one lab.

stack

  • surface electromyography
  • video
  • robot arms and grippers

hiring

  • Robotics Software Engineer
  • AI Research Engineer
  • Research Engineer, Human Biomechanics
  • Hardware Engineer
  • Data Operations Lead

pricing

model: unknown

motion

unknown

channels

  • LinkedIn
  • research hiring

confidence

low

Founders

  • name: Yi Zhang
    title: Founder
    role: Founder
    bio: Co-Founder & CEO of Relari. Previously building autonomous cars & trucks at Pony.ai and warehouse robots at Dexterity.ai. Harvard MBA.
  • name: Pasquale Antonante
    title: Founder
    role: Founder
    bio: Co-Founder & CTO of Relari. I have a PhD from MIT focusing on the reliability of complex AI systems such as autonomous cars. Previously at NVIDIA and Raytheon Technologies (previously United Technologies).

Back to YC companies