Sereact / Cortex 2.0
- hybrid
- 2026
- WAM + VLA
Model cardPaper ↗
Cortex 2.0 is Sereact's industrial policy with a world-model planning loop: it scores imagined visual futures before it moves.
Key facts
Rows below are the register record in lib/models.ts. A dash means the value is not confirmed from a primary source. Classification follows Reuss, NVIDIA Technical Blog (15 June 2026) and the model's own paper.
| Name | Cortex 2.0 |
|---|---|
| Lab | Sereact |
| Year | 2026 |
| Paradigm | hybrid |
| Backbone | — |
| Source | https://arxiv.org/abs/2604.20246 |
What the paper claims
- The system extends Sereact's Cortex VLA with a world model that generates candidate future trajectories in visual latent space. A Process-Reward Operator scores each candidate for progress, risk and completion; the policy commits to the highest-scoring rollout. Aida et al., 2026
- Because planning is in visual space, the authors say the same loop transfers across single-arm, dual-arm and humanoid embodiments. Training is grounded in data from Sereact's operating warehouse fleet. Aida et al., 2026
What it evaluates
- They benchmark four real-world industrial tasks of increasing complexity: pick-and-place, item and trash sorting, screw sorting, and shoebox unpacking, on single-arm and dual-arm platforms. Aida et al., 2026
- They report the best success rates among the baselines they run, success rates beyond 90% on all four tasks, 99% after continued in-domain operation, and zero human interventions across those rollouts. Baselines in the paper include π0.5, Diffusion Policy and RDT-2. Aida et al., 2026
On this register
Hybrid. Mixes the vision-language-action (VLA) and world action model recipes. That axis is defined in what is a world action model? and in Reuss, NVIDIA Technical Blog (15 June 2026).
Other hybrid models on the register: Motus, BagelVLA.
The backbone field is unconfirmed, so this card does not claim a video-model family.
Coverage
No stories on this model yet.