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.
Reality
- Evidence40
- Adoption15
- Hype gap+35
- Incentives70
- Confidence45
AMD agreed to buy Fei-Fei Li's World Labs for about $8.2 billion in stock as GPU suppliers move into robotics model work. Robotics teams now have to decide how much of their data and failed experiments a compute vendor gets to see.
Reality
- Evidence55
- Adoption
- Not scored
- Hype gap+30
- Incentives70
- Confidence60
The rally count, the autonomy claim and the model's identity all come from one press release. The machine that fell over and got up cannot be the legless robot Galbot actually sells.
Reality
- Evidence25
- Adoption20
- Hype gap+55
- Incentives70
- Confidence55
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.
Reality
- Evidence35
- Adoption20
- Hype gap+30
- Incentives75
- Confidence55
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.
Reality
- Evidence33
- Adoption
- Not scored
- Hype gap+32
- Incentives78
- Confidence42