MindGrip AI · SO-101 LabLab overlay for imitation-learning robot arms — SO-101 and beyond.Acronyms & short names
Glossary for MindGrip AI / so101-lab and the SO-101 pick-and-place sorting path. Use this when you see short forms like HF, ACT, FK, or VLA without expansion.
Tip: keep this open beside the project page or roadmap.
How to use this page
| Column | Meaning |
|---|---|
| Acronym | Short form as written in the docs |
| Stands for / full name | Expanded meaning |
| In this lab | Why it matters for MindGrip / so101-lab |
Platforms, products & this repo
| Acronym | Stands for / full name | In this lab |
|---|---|---|
| MindGrip AI | MindGrip AI (project brand) | Cognitive targeting / handling narrative for the binary sorter lab. |
| so101-lab | This overlay repository | Plugins, CLI menu, patches, and docs — sits beside vanilla lerobot/, not a fork of it. |
| HF | Hugging Face | Hosts models, datasets, Spaces, and HF Jobs (cloud GPUs). |
| Hub | Hugging Face Hub | Where you push datasets and publish / pull policies (e.g. SUBHENDU/…). |
| HF Jobs | Hugging Face Jobs | Managed cloud GPU training — preferred for Step 8 from LeLab (a10g-small / a10g-large). |
| LeRobot | LeRobot (Hugging Face robotics library) | Core framework: teleop, datasets, train, rollout. Upstream docs: huggingface.co/docs/lerobot. |
| LeLab | LeRobot Lab (GUI) | Recommended GUI for calibrate → teleop → record → train → deploy on a laptop/desktop. |
| CLI | Command-Line Interface | Terminal path: ./so101_cli_menu.sh and uv run lerobot-*. |
| GUI | Graphical User Interface | LeLab (and Rerun) instead of long flags. |
| uv | uv (Astral package / project manager) | Installs deps and runs tools (uv run …, uv sync). |
| TTY | Teletype / terminal device | Interactive stdin used by recording key patches on macOS. |
| JSON | JavaScript Object Notation | Calibration files — this lab: LeoArm.json under so_leader/ and so_follower/. |
| ENV | Environment / config file | .so101_cli.env — ports, ids, camera, TASK, dataset (menu C). |
| MJPG | Motion JPEG | Preferred USB camera fourcc for stable capture. |
Hardware & robotics
| Acronym | Stands for / full name | In this lab |
|---|---|---|
| SO-101 / SO-ARM101 | Standard Open Arm 101 | Leader + follower dual-arm kit for teleop and sorting. |
| SO-100 | Standard Open Arm 100 | Earlier related arm; mentioned next to SO-101 in some upstream docs. |
| DOF | Degrees Of Freedom | SO-101 is 6-DOF (six actuated joints including gripper). |
| STS3215 | Feetech STS3215 servo | Smart servo on the arms (bus IDs 1–6 per arm). |
| TTL | Transistor–Transistor Logic (serial bus) | Feetech motor bus (~1 Mbps). |
| USB | Universal Serial Bus | Arm adapters (/dev/tty.usbmodem… on macOS) and the policy webcam. |
| PSU | Power Supply Unit | External 5 V (leader) and 12 V (follower) — USB alone is not enough. |
| ID | Identifier | Motor bus ID (1–6) and robot/teleop id — this lab uses LeoArm for both arms. |
| J1–J6 | Joint 1 through Joint 6 | Pan, lift, elbow, wrist flex, wrist roll, gripper. |
| EE | End-Effector | Gripper / tip frame that picks objects. |
| FK | Forward Kinematics | Gripper pose from joint angles. See FK & IK. |
| IK | Inverse Kinematics | Joint angles from a desired pose (Jacobian / DLS demos on the kinematics page). |
| DH | Denavit–Hartenberg | Classic link parameters used when writing each joint’s 4×4 transform. |
| L<sub>max</sub> | Maximum reach length | ≈ 38 cm for SO-101 — size the sorting mat inside this arc. |
| URDF | Unified Robot Description Format | Robot model for sim / kinematics helpers. |
| MJCF | MuJoCo XML model format | Used by the optional so101_mujoco plugin. |
| ROS2 | Robot Operating System 2 | Classical robotics middleware — not part of this overlay’s install or CLI. |
| RViz | ROS Visualization | ROS 3D viewer — not shipped or documented as a MindGrip path. |
Vision, sensing & control
| Acronym | Stands for / full name | In this lab |
|---|---|---|
| OpenCV | Open Source Computer Vision Library | Backend for the USB cam (type: opencv). |
| FPS | Frames Per Second | Policy cam: 640×480 @ 30 FPS, name front. |
| Hz | Hertz | Control / teleop loop rate (often ~30–60 Hz). |
| Mbps | Megabits per second | Motor bus baud (~1 Mbps). |
| Rerun | Rerun (visualization SDK) | Live camera + joint plots when --display_data=true / CLI viz. |
| OBS | Open Broadcaster Software | Virtual camera path covered by an optional patch. |
Learning, policies & ML
| Acronym | Stands for / full name | In this lab |
|---|---|---|
| IL | Imitation Learning | Learn from teleop demos (not hand-coded pick scripts). |
| BC | Behaviour Cloning | Map observations → actions from demos. |
| ML | Machine Learning | Training the neural policy (Step 8). |
| Policy | Control policy (neural net) | Maps camera + joints → motor commands. |
| ACT | Action Chunking Transformer | Recommended first policy for this sorter. |
| Diffusion | Diffusion Policy | Smoother multi-modal actions; heavier than ACT. |
| VLA | Vision–Language–Action | Policies that take images + language (SmolVLA, Pi-0, …). |
| SmolVLA | Small VLA (LeRobot) | Advanced option; train on cloud GPUs only. |
| Pi-0 / π₀ | Physical Intelligence π₀ | Large flow-matching VLA — cloud / big GPU. |
| DiT | Diffusion Transformer | Backbone used in some multitask policies. |
| RL | Reinforcement Learning | Reward/trial-and-error — separate from this IL path. |
| OOM | Out Of Memory | Risk when training large policies on 8–16 GB hosts. |
| VRAM | GPU memory | On NVIDIA cloud cards; Apple Silicon uses unified memory. |
Compute & hosts
| Acronym | Stands for / full name | In this lab |
|---|---|---|
| CPU | Central Processing Unit | Teleop, record, light inference on laptop/desktop or Pi. |
| GPU | Graphics Processing Unit | Prefer cloud NVIDIA for ACT training (HF Jobs / AWS). |
| MPS | Metal Performance Shaders | Apple Silicon PyTorch backend — tiny local experiments only. |
| CUDA | NVIDIA GPU compute API | Required for serious lerobot-train (CLI · 12 / 13). |
| Pi | Raspberry Pi | Valid control host for teleop/record/rollout — not for comfortable local training. |
| SSD | Solid-State Drive | Keep free space for demos and checkpoints. |
| GB | Gigabyte | Host RAM guidance: 8–16 GB laptop/desktop class. |
Quick “say it out loud” cheat sheet
| You see… | Read it as… |
|---|---|
| HF Jobs | Hugging Face Jobs |
| ACT | Action Chunking Transformer |
| FK / IK | Forward / Inverse Kinematics |
| L<sub>max</sub> | Max reach (~38 cm) |
| VLA | Vision–Language–Action |
| DOF | Degrees of Freedom |
| CLI / GUI | Command-line / graphical interface |
| PSU | Power supply unit |
| IL | Imitation learning |
| LeoArm | Calibration / robot id for both arms in this lab |
Related pages
- MindGrip AI Home — what this overlay is for
- Pick & Place Sorting — project overview
- Roadmap — 9-step build path
- LeLab · CLI menu — GUI vs terminal
- Forward & Inverse Kinematics — FK / IK / L<sub>max</sub>
- Policy Cheatsheet — ACT vs Diffusion / VLA
- Directory Structure — so101-lab layout