Today, we’re announcing Axis V2 — the compounding data engine for Physical AI. Axis V2 delivers scalable online post-training to the Axis data engine, solving Physical AI’s biggest bottlenecks: scale, diversity, and iteration speed. The core paradigm shift is straightforward. Rather than relying solely on pre-defined data collection workflows, a trained policy becomes an active component of the data engine — its failure cases dictate what data is collected next. Policy rollout → Failure states → Targeted tasks → Human-gated corrections → Policy retraining → Enhanced policy → New rollout ♻️ Failures translate into targeted tasks; human corrections power the next iteration of policy training, and the refined policy uncovers the next set of required data. Data and policy now evolve in lockstep within an uninterrupted feedback loop. Full details are outlined below ⬇️