Last updated:
Dynamic intelligence Introduces Fleet-Tape v0: Production Telemetry and Policy Promotion for Embodied AI Fleets
Fleet-Tape v0 is Dynamic intelligence’s production telemetry and promotion layer for embodied AI fleets—so checkpoint history, staged rollouts, and replay exports stay tied to what robots actually did on the floor.
Why fleet truth is a different problem than training logs
Ground-Log gives you governed episodes for learning. Fleet-Tape is what you need once a policy is live: continuous telemetry, policy identity, and promotion/rollback hooks that survive Tuesday’s layout change, vendor patch, or SKU mix.
What Fleet-Tape v0 includes
• Policy lineage: Immutable hashes and version tags for checkpoints in production, mapped to robots, zones, and time windows. • Promotion and rollback: Staged fleet percentages, automatic halt signals, and one-click rollback to a last-known-good checkpoint. • Replay exports: Slice production traffic for root-cause review, insurer bundles, or Bench-Fabric regression—without copying raw lakes by hand.
How it pairs with Ground-Log and Bench-Fabric
Fleet-Tape closes the loop: incidents in production become searchable tapes; Ground-Log captures the multimodal slice you need for fine-tuning; Bench-Fabric runs scenarios before the next promotion. None of the three replaces a research VLA—they make shipping one defensible.
Who should adopt first
Operators running multiple sites, mixed OEM stacks, or human–robot shared workspaces where “it worked in sim” is no longer an acceptable incident post-mortem.
Early access
Design partners get Hub views for fleet-wide policy state, rollout templates, and engineering playbooks to wire OEM telemetry and CI gates in parallel with Ground-Log.