New implementation of the QR robot planning/simulation/control stack.

NQR depends on sibling checkouts of HPN and QAA (editable pixi path dependencies, used by the planning/policy stack). Clone all three side by side first — pixi install fails with error while canonicalization ../HPN otherwise:

GitHub/
├── NQR/    (this repo)
├── HPN/
└── QAA/

Then:

pixi install   # solves the env and builds the qr_mapping C++ module
pixi run test

Notes:

• Supported platforms: macOS arm64 and Linux x86-64 (Ubuntu jammy/noble).
• The first install downloads the Drake C++ SDK (~600 MB, cached under .pixi/fetchcontent) and compiles qr_mapping; expect several minutes.
• src/qr_belief/, src/qr_perception/, src/qr_policy/, src/qr_main/, src/qr_run/ host the belief/perception/policy stack; planning itself (RetroPlan, TAMP domains, Roboverse) is imported from the HPN checkout.

• src/qr_api/ — interfaces and typing for the pipeline components.
• src/qr_robots/ — virtual robots (RBY1, Spot) backed by MuJoCo and Drake simulation servers over a shared ZMQ protocol.
• src/qr_kinsim/ — kinematics-only simulator (Pinocchio + Coal + Meshcat) with object attachment for grasp/push modeling.
• assets/ — robot models and test scenes.

# RBY1 pick-and-carry in the kinematic sim (open the printed Meshcat URL)
pixi run kinsim-demo

# A MuJoCo sim server with an interactive viewer
pixi run python -m qr_robots.mujoco.rby1.sim --mode display \
    --model-dir assets/test_scene \
    --objects '{"table": {"file": "table.xml", "fixed": true}, "block": "block.xml"}'