A practical public notebook for building affordable robots, choosing an open software stack, tracking robotics patents by subsystem, and turning research into testable prototypes without pretending a patent search grants permission to practice protected claims.
The linked Short could not be programmatically verified from this build node, so this site does not invent the exact model. These three paths cover the most common low-cost embodiments and let the software/intelligence stack stay reusable.
3D-printed structure, serial-bus servos, USB servo controller, 12 V supply, gripper, camera, and a Linux or Mac development machine.
*Reference range from the open AM-ARM project for follower / leader+follower configurations; sourcing changes over time.
AM-ARM reference ↗Six bus servos per arm, servo adapters, 3D-printed components, camera(s), power supplies, and LeRobot for teleoperation, dataset capture, and learned policies.
Official SO-101 guide ↗A research-scale biped using 3D-printed parts, off-the-shelf components, control/runtime tooling, identification, simulation and training environments.
*Published 2026 project estimate; shipping, taxes and sourcing can change the total.
LeRobot Humanoid ↗actuatorsmotor/servo busregulated PSUfuseE-stopUSB/serial adaptercameraIMU optional3D-printed/CNC framefasteners + bearings
For a first prototype, prioritize reliable joint sensing and safe power distribution over expensive AI compute.
Mac: development, dataset review, UI, model experimentation.
Robot-side Linux: real-time-ish device access, ROS 2 nodes, ros2_control, cameras, motor buses and safety interlocks.
Optional accelerator: only add a GPU/edge board when your vision or policy model actually needs it.
Messaging, nodes, transforms, services and ecosystem integration.
Docs ↗Hardware interfaces, controllers and a clean boundary between robot devices and higher-level software.
Docs ↗Python-native real-robot control, synchronized datasets, imitation learning and vision-language-action policies.
Docs ↗Motion planning, collision checking and manipulator workflows.
Docs ↗Simulation before risking hardware. Validate geometry, sensors and control loops first.
Docs ↗Low-cost visual perception, calibration, marker tracking and prototyping.
OpenCV ↗Use patent publications as architecture signals: identify the problem, inputs, control loop, sensing, actuation, verification and failure handling. Preserve the publication number and assignee. Then independently design your implementation and run a real freedom-to-operate review before commercialization.
Assignee: Amazon Technologies, Inc.
Research lesson: treat storage geometry, end-effector choice, perception and pick verification as one coordinated task pipeline.
Assignee: Collaborative Robotics
Research lesson: study mobile-manipulator embodiment, locomotion/manipulation coordination and human-shared workspace constraints.
Assignee: listed inventor publication record
Research lesson: separate perception, learned representation, decision policy and action validation so each can be tested independently.
Assignee: Marvell Asia Pte Ltd
Research lesson: robotics performance can be limited by arithmetic/dataflow latency; parallel accumulation is an architecture topic, not just a model topic.
Assignee: Apple Inc.
Research lesson: mechanical preload, bearings, sensing and actuator stability belong in the robotics intelligence map because precision depends on mechanics.
B25J manipulatorsB62D57 walking/mobile structuresB65G47/90-91 pick/placeG05D controlG06V visionG06N machine learning
Use these as search lanes, then follow citations and patent families to widen coverage.
USPTO B25J scheme ↗| Field | Store | Why |
|---|---|---|
publication_id | Patent/publication number | Exact provenance |
family_id | Family / related filings | Avoid treating one filing as the whole invention family |
assignee | Named record owner at publication | Source context |
subsystem | Manipulation / mobility / sensing / control / compute / HRI | Searchable ontology |
architecture_lesson | High-level generalized insight | Reusable learning without copying claim language |
evidence_links | USPTO / paper / repo / datasheet | Verification |
ip_status | Research-only / review-needed | Prevents “published = free to use” mistakes |
prototype_test | Simulation / bench / low-speed hardware | Convert reading into measured engineering |
Mechanical stop clearance • correct voltage/polarity • current-limited bench testing • physical E-stop • low torque/speed limits • no loose clothing/cables near joints • stable base.
Joint limits • watchdog timeout • comms-loss stop behavior • collision envelope • supervised test area • logging • reproducible reset/recovery procedure.
ROS 2 docs · ros2_control · LeRobot · MoveIt 2 · Gazebo
USPTO Patent Public Search · USPTO CPC B25J · Render static hosting · GitHub Pages