← Robotics map

Rhoda AI

FutureVision: a robot brain pre-trained on internet video to predict how the physical world moves

Scaleup · Pilot · Robot foundation model · Palo Alto, California, United States · Founded 2024 · rhoda.ai

Profile

Rhoda AI left stealth in Mar 2026 with FutureVision, a robot foundation model that learns how objects move from hundreds of millions of internet videos and then controls robots in closed loop. It runs on Rhoda's own systems in industrial pilots and is meant to work with many kinds of robot hardware. CEO Jagdeep Singh previously founded QuantumScape. It raised a $450M Series A at a $1.7B valuation.

Get weekly updates on Rhoda AI and its peers

The week in humanoid robotics, funding and AI, with the sources linked. Free, in your inbox.

One email a week. Free. Unsubscribe any time.

People

As of . X handles are shown only when we could confirm they belong to the person.

Appearances

Talks, panels and interviews by this company’s leaders, newest first. What a speaker said is their own claim, not our finding.

  1. USPublished
    Podcast

    Automated with Brian Heater (A3 / automate.org)

    “Jagdeep Singh: Teaching Robots From Internet Video at Rhoda AI”

    Jagdeep Singh, Co-founder & CEO

    Summary (From the transcript)

    Singh said Rhoda's model, pre-trained on hundreds of millions of video clips, learns a new task from 10 to 20 hours of robot data, and relearned a container-folding task at a factory with about three hours of extra data. He said Rhoda plans to ship its model on off-the-shelf hardware very soon while it builds its own wheeled mobile manipulator, including its own actuators, tactile sensing and hands. He also said customers must give Rhoda a perpetual license to train on data from their sites.

    What Jagdeep said (the speaker’s own claims, not our findings)

    • New tasks need 10-20 hours of robot data after video pre-training; a factory container-folding task was relearned with about 3 hours of factory data (Singh's claim).
    • Rhoda plans to deploy its model on off-the-shelf hardware very soon and thinks it could be the first autonomous manipulation-capable robot deployed at scale (Singh's claim; no date given).
    • Its own hardware targets about 45 kg payload close to the body and 18-19 kg far from it (about double the OSHA guideline), three years of continuous industrial operation, and CE-certifiable subsystems (Singh's claim).
    • The robot is a wheeled mobile manipulator with 7-degree-of-freedom arms; Rhoda is building its own actuators and tactile sensing and weighing three-finger hands (Singh's claim).
    • Rhoda bought third-party humanoids early on and burned out one motor within a day and another within a week (Singh's claim).
    • Customers must grant an irrevocable, perpetual license for Rhoda to train on data from their facilities (Singh's claim).
    • Rhoda has spoken with 100+ customers and has not seen a next-gen autonomous manipulation robot doing work in the field (Singh's claim).
    • Current tasks include decanting ball bearings for an automotive customer and folding plastic containers at another company (Singh's claim; customers not named).

As of . Items we could not verify are marked or left out.

Flagship products

Funding and valuation

Hiring signals

Open roles
43all roles
Top location
Mountain View · 95%41 of 43 roles
Top role area
AI & foundation models10 roles · 24% of 42 engineering & ops

Trademarks

3 names filed in the US since 2025 · last checked .

Source: USPTO Trademark Search, matched on the company’s legal names. Rechecked on the 1st and 15th of each month. A filing shows intent to use a name in the US, not a launched product.

Competitors

Companies with overlapping products. Direct = same customers and same job; Adjacent = overlapping technology or market, different focus.

Show 5 more

Recent news

Sources

Confidence: high.

Suggest a correction

Get the weekly email briefing

One email a week. Free. Unsubscribe any time.