05 · Roadmap

Gaps — framed honestly

Strong sim + modular software; a real-robot product still needs hardware, vision, and trained skills. This is the v0.3.0 launch snapshot.

Done in repo v0.3.0

Modular stack + World ModelDone
Mock + MuJoCo simDone
CLI (interactive + demos) + demo GIFDone
Language (rule / BitNet / HTTP)Done
Planner, policy, safetyDone
ROS2 adapter (dry-run)Done
Vision / tracker interfacesDone
Benchmarks + CI + DockerDone
CONTRIBUTING, issues, CHANGELOG, release notesDone

Still open

P0Live robot drivers (ROS2/SDK), calibration
P0Production detectors — pretrained YOLO adapter done (scene.backend: yolo, optional ultralytics); depth-based 3D pose still open
P1Diffusion policies + LIBERO-style eval (BC-trained MLP on randomized mock pick: done, kinetic train-bc — trained 100% vs random-init 0%)
P1Real IK — numeric DLS backend done (control.ik_backend: dls); MoveIt/pinocchio integration for real arms still open
P2Strong instruct LLM default; mobile base nav
P2Edge deploy (Jetson / TensorRT)
P3Safety certification, multi-arm / humanoid

Summary

Strong sim + modular software; real-robot product needs hardware, vision, and trained skills.

This roadmap is the honest gaps list for Kinetic. Questions or pilots: contact DifrasAI.