Skip to content

Topic

Robot Learning

Learned, data-driven control policies replacing hand-engineered perception-planning-control pipelines in robotics.

Current stories

buildOne report1 publisher

Odyssey opens its Odyssey-3 world model to free public testing

Odyssey opened a free public research preview of Odyssey-3, a world model that generates interactive environments from text prompts in real time. The demo runs the Flash variant, while the 66.1 physics score Odyssey cites came from its Pro model on a best-of-eight run.

Publishers:the-decoder.com

Reality

Evidence40
Adoption15
Hype gap+35
Incentives70
Confidence45
investOne report1 publisher

Figure commits $3.5bn of compute to a humanoid market that shipped 19,100 units

Figure's software finished 56% of chores in 30 unfamiliar homes, Boston Dynamics began training Atlas inside a Hyundai plant, and Unitree priced a 22-joint hand at $6,500.

Reality

Evidence34
Adoption42
Hype gap+38
Incentives72
Confidence44
buildOne report1 publisher

Odyssey's simulation-trained policy went 77% as far between interventions as its real-footage policy

Odyssey-3 keeps one visual world model frozen and trains small action decoders on top of it for arms, humanoids, drones, cars and game agents. Driving is the only test it ran in closed loop on public roads.

Publishers:runtimewire.com

Reality

Evidence35
Adoption20
Hype gap+30
Incentives75
Confidence55
productOne report1 publisher

Vision plus touch gets a robot hand to 85% success across five tasks and 25 objects

Humanoid locomotion is rehearsed and it films well. The manipulation figure in the published record is 85% success on five tasks with 25 objects. A household or warehouse pilot has to clear that bar on its own objects.

Reality

Evidence46
Adoption20
Hype gap+26
Incentives54
Confidence55
buildOne report1 publisher

Decoupling the policy from the control loop is the load-bearing part of LeRobot

The paper argues that fragmented middleware and dataset formats slow robot learning more than algorithms do, and its asynchronous inference stack is where that argument becomes something you can test on your own hardware.

Publishers:arxiv.org

Reality

Evidence33
Adoption
Not scored
Hype gap+32
Incentives78
Confidence42