Kinetic

See. Think. Move.
Lightweight modular Vision–Language–Action for robotics.

Quick start · What this is / isn’t · Gaps · Changelog · Releases

Not one giant transformer: specialized modules talk through a World Model.

Instruction → Language → Planner → World Model
Vision / sensors → Scene → World Model
World Model + subtask → Policy → Control → Robot
                 ↑____________ feedback ___________|

Quick start

Requires Python 3.10+ (CI and primary dev: 3.11).

git clone https://github.com/AshrafGalibShaik/KINETIC.git
cd KINETIC
pip install -e ".[dev,sim]"

# Hero path (60 seconds)
kinetic demo pick          # always works (mock)
# or physics + 3D window:
kinetic demo mujoco        # needs MuJoCo + display

# Interactive CLI (B/W retro-futuristic)
kinetic

# More
kinetic demo list
kinetic run -i "Pick up the bottle"
kinetic train-bc            # behavior cloning: demos → MLP → eval (100% vs 0% random)
kinetic eval --seeds 5      # multi-task success rates (pick-bottle, place-cup)
kinetic vla-bench --trials 5
pytest -q

Optional backends via config: control.ik_backend: dls (Jacobian damped-least-squares IK), scene.backend: yolo (pretrained detector, pip install ultralytics), policy.backend: lerobot (run pretrained HF VLA checkpoints — SmolVLA, ACT — through the loop; pip install "lerobot[smolvla]"; CPU works: ~27s per 50-action chunk).

What this is / isn’t

This is	This is not
Modular VLA software stack (sim-first)	A single 7B end-to-end OpenVLA clone
Event-driven language + world model	LLM joint commands every control tick
Mock + MuJoCo demos you can run today	Turnkey industrial robot product
Interfaces for ROS2 / real arms	Certified safety controller
Latency + literature VLA comparison	LIBERO leaderboard without trained policies
BC-trained sim policy + multi-task eval	A frontier foundation model (yet)

See docs/gaps.md for the honest roadmap.

Support matrix

	Status
OS	Windows 10/11 (primary), Ubuntu (CI)
Python	3.10+ · 3.11 recommended / CI
Mock sim	Always
MuJoCo	Optional: pip install -e ".[sim]"
BitNet LLM	Optional local install (docs/bitnet.md)
ROS2 arm	Dry-run adapter; live when rclpy + config
Real Franka/UR5	Adapter pattern ready; drivers not bundled

Repository layout

KINETIC/
├── README.md · CHANGELOG.md · CONTRIBUTING.md
├── LICENSE · NOTICE · Dockerfile · pyproject.toml
├── configs/          # default · mujoco · bitnet
├── assets/mujoco/    # tabletop scene
├── docs/             # architecture · gaps · bitnet · benchmarks · assets
├── scripts/          # BitNet setup · demo GIF
├── src/kinetic/      # package
└── tests/

Module	Role
schemas	World / task / action models
vision · scene · world	Perceive + persistent state
language · planner	Intent + subtasks
policy · control · robots	Act + safety + adapters
loop · factory	Wire + run
ui · demos · bench	CLI + demos + benchmarks

Configs

File	Use
configs/default.yaml	Mock sim (CI / fast)
configs/mujoco.yaml	Physics + viewer
configs/bitnet.yaml	Optional 1.58-bit reasoner

Docs

Doc
Architecture	Design
Gaps	What’s left
BitNet	Optional LLM
Benchmarks	vs lightweight VLAs
Contributing	PRs / setup
Release notes v0.3.0	Launch notes

Robots

Backend	Status
mock_sim	Default lightweight arm
mujoco	Tabletop physics + cameras
ros2	Joint trajectory (dry-run by default)

New arms: implement robots.base.RobotAdapter.

Design principles

1. World Model is the integration bus
2. Language never emits joint commands
3. Safety lives in the motion controller
4. LLM / planner run on events, not every tick

License

Apache-2.0 — LICENSE · NOTICE.