🗺️ Roadmap

Checked items are available in V1. Unchecked items are planned work; current robot and device support is listed in compatibility.

Available

• Policy learning, learned diffusion and reusable priors
• JAX planning and composable constraints
• Six task families and planner baselines
• Configured experiments, metrics, traces and replay

Next

• MGA CUDA qualification and humanoid walking/push
• Torch planner parity and device-resident array transfers
• Rust/C++ control and I/O components
• Panda/xArm7 deployment controllers and ros2_control actions
• Async planning/control, command expiry and fault recovery
• Reproducible hardware latency, reliability and task-quality benchmarks

Explore

• MuJoCo Warp rollout integration
• Action-chunk/VLA proposals through OpenPI and LeRobot adapters
• Live perception, timestamped geometry and dynamic scenes

Backend acceptance gates · Benchmark contract