Annin Robotics Community

  • John Dammeyer@johndammeyer· 198 viewsin Robot Builds

    The table top is done and screwed to the cabinet.  The AR3 bolted in place.  I found I needed a dedicated PC for it so I grabbed an old one that had LinuxCNC and installed WIN-10 and put what I hoped was the current working version of the AR4 software on it.  

    Well yes and no.   I had forgotten I'd converted the encoders and steppers to match the AR4 but the rest of the system is still an AR3.   Couldn't remember what firmware was in the Teensy 4.1.  

    Finally got it all working without stepper/encoder issues while jogging.

    Loading comments…

    Loading comments…
  • Santamaria@santamaria· 244 viewsin Robot BuildsFlorida Institute of Technology

    Cobot Autonomous Living Interface C.A.L.I is complete and able to help feed a person fully autonomously

    Loading comments…

    Loading comments…
  • Paolo Gagge@gagge· 529 viewsin Robot Builds

    PaMiLoCELL – Project Development #MilestonesFrom robotic arm assembly to AI-powered material identification and sorting.

    The PaMiLoCELL project is structured into five progressive development milestones:

    #Milestone 01 – Robotic Arm & Basic Cell Assembly

    • Complete the Annin Robotics AR4-MK5 mechanical assembly.
    • Build the basic robotic cell structure.
    • Install and configure the overhead 2D camera.
    • Perform preliminary mechanical and electrical checks.

    #Milestone 02 – AR4 Software & Motion Control

    • Install and explore the latest AR4 control software.
    • Learn robot programming, kinematics and motion control.
    • Perform controlled tests of individual joints and basic robot movements.
    • Verify positioning accuracy and repeatability.

    #Milestone 03 – Electrical Control Panel & Functional Safety

    • Assemble and wire the electrical control panel.
    • Integrate the PLC, power supplies and control components.
    • Install emergency stops, safety interlocks and safety relay.
    • Verify and validate safety functions before automated operation.

    #Milestone 04 – PLC Programming & Cell Automation

    • Learn and configure the PLC programming environment.
    • Develop control logic and operational sequences.
    • Establish communication between PLC, robot controller and peripheral devices.
    • Test interlocks, fault management and automated sequences.

    #Milestone 05 – AI Vision, Sensor Fusion & Intelligent Sorting

    • Integrate the 3D camera and additional material-identification sensors.
    • Develop three interconnected functional layers:
    • Sensory Layer: 2D/3D vision, data acquisition and material sensing.
    • Cognitive Layer: AI-based object recognition, material classification and decision-making.
    • Actuation Layer: Robot motion planning, gripping and sorting execution.
    • Integrate and validate the complete perception-to-action workflow.

    #First Application – Material Identification & Sorting

    The initial objective is to demonstrate the automated identification, classification, picking and sorting of small waste objects using a robotic cell integrating industrial automation, AI vision and sensor fusion.

    #Development approach: Build → Learn → Integrate → Test → Validate.

    PaMiLoCELL | An independent technical development project by PaMiLoLAB.

    #PaMiLoCELL #PaMiLoLAB #AnninRobotics #AR4MK5 #Robotics #IndustrialAutomation #PLC #MachineVision #ArtificialIntelligence #WasteSorting

  • Paolo Gagge@gagge· 505 viewsin Robot Builds

    🔧 PaMiLoCELL #11 | J4 Timing Hub – Alignment & Bearing Resistance

    Can a 0.15 mm deviation cause a robotic joint to bind?

    During assembly of the Annin Robotics AR4-MK5, I noticed resistance over ~30° of rotation at J4. Loosening the J4 Timing Hub made it disappear.

    📏 1. What I measured

    • Required hub offset: 25.30 mm

    • Measured: 25.15–25.21 mm

    • Hub difficult to slide onto the shaft while keeping the pulley face square (90°).

    A 0.15 mm offset alone is unlikely to explain the issue. More likely causes: an out-of-square hub, a tight/uneven bore, or excessive axial bearing preload.

    🔍 2. Technical feedback from Community

    • Keep the bearing contact face flat and parallel: avoid aggressive Dremel machining.

    • If the bore is tight, lightly finish it without introducing play.

    • Check printed hub thickness: ≤15 mm at the flat face.

    • Tighten preload screws only until bearing play is removed and rotation remains smooth.

    • Reprint the hub if its bore or face is distorted.

    ⚠️ 3. A disassembly mistake

    I overlooked the different set-screw lengths:

    • 1 × M4×10 engages the shaft alignment hole.

    • 3 × M4×5 secure the hub around its perimeter.

    • 4 × M4×5 face screws adjust bearing preload.

    During removal, the longer screw was still engaged, damaging both the printed hub and the aluminum shaft.

    🧱 4. Printed polymer vs. aluminum

    The hub is designed to be 3D-printed (PETG or equivalent; 50% infill, 4 walls).

    Aluminum adds machining cost without a demonstrated performance benefit for this application.

    ✅ 5. Next steps

    Inspect the shaft; reprint the hub if needed; check the 25.3 mm position and squareness; set light preload; verify smooth, play-free rotation.

    💡 LESSON LEARNED

    A dimensional deviation is not automatically the root cause.

    Check alignment, bore fit, bearing preload and screw lengths before modifying a component.

    Thanks to the Community for the technical exchange!

    ❓ Repair or reprint: what criteria guide your decision?

    🔧 6. Additional Assembly Issues Resolved

    ✅ Corrected blind or incompletely drilled holes.

    ✅ Eliminated minor belt and pulley interference points, ensuring adequate mechanical clearance for safe operation.

  • Paolo Gagge@gagge· 624 viewsin Robot Builds

    PaMiLoCELL #10 – 10 Assembly Lessons Learned | AR4-MK4 → MK5

    Building a robotic arm is not just a sequence of assembly steps. It is a series of small engineering decisions that affect reliability, serviceability and repeatability.

    Here are 10 lessons from recent hands-on work:

    1️⃣ SOLDERING & WETTING

    Solder should wet the heated pin and conductor—not simply melt onto the iron. My trials with Sn99.3Cu0.7 and Sn63Pb37 reinforced the importance of surface preparation, flux and heat transfer. Tip temperature alone is not a quality criterion.

    2️⃣ FASTENER DIMENSIONS

    An M4×14 countersunk screw is measured including its head; an M3×16 socket-head screw is measured below the head. Geometry matters as much as nominal length.

    3️⃣ DRILLING, TAPPING & FIT

    For a conventional M4×0.7 cut thread, a Ø3.3 mm pilot hole is typical. Tap alignment, deburring, engagement depth and clearance all need checking. Never use tightening torque to force misaligned parts together.

    4️⃣ PULLEYS & MATERIAL SELECTION

    A timing hub transfers torque. Aluminium versus engineering polymer must be evaluated against loading, shaft fit, set-screw retention and dimensional stability—not appearance alone.

    5️⃣ BACKLASH

    Lost motion during reversal is not identical to belt elasticity or friction. Adjust tension without introducing excessive preload, then check smooth motion over the full range.

    6️⃣ CABLE PROTECTION

    Braided sleeving and cloth harness tape serve different functions. Abrasion, bend radius, strain relief and movement determine the correct choice.

    7️⃣ 3-PIN CONNECTORS

    For Hall sensor wiring, verify VCC, GND and signal against the actual pinout. Identical black wires and reversible-looking connectors demand clear identification.

    8️⃣ J2 ENCODER GND

    The 2×12 terminal block does not allow the A/B electrical connection to be identified by row position alone. Follow the manual, then verify electrically.

    9️⃣ CONTINUITY TESTING

    With all power sources and USB disconnected, I verified continuity between the J2 encoder GND terminal, terminal B/GND and Teensy 4.1 GND.

    The multimeter BEEP confirmed continuity—not near-zero resistance or functional encoder performance.

    🔟 INCOMING QUALITY & COST

    Inspect real screw dimensions, corrosion and threadlocker residue before installation. For imported MK5 components, compare the total landed cost—not just the purchase price.

    🎯 Main takeaway:

    Assembly quality is built through measurement, inspection and controlled verification. Functional validation comes next.

    #PaMiLoCELL https://www.linkedin.com/search/results/all/?keywords=%23pamilocell&origin=HASH_TAG_FROM_FEED #PaMiLoLAB https://www.linkedin.com/search/results/all/?keywords=%23pamilolab&origin=HASH_TAG_FROM_FEED #AR4Robot https://www.linkedin.com/search/results/all/?keywords=%23ar4robot&origin=HASH_TAG_FROM_FEED #Robotics https://www.linkedin.com/search/results/all/?keywords=%23robotics&origin=HASH_TAG_FROM_FEED #MechanicalEngineering https://www.linkedin.com/search/results/all/?keywords=%23mechanicalengineering&origin=HASH_TAG_FROM_FEED #Electronics https://www.linkedin.com/search/results/all/?keywords=%23electronics&origin=HASH_TAG_FROM_FEED #QualityEngineering https://www.linkedin.com/search/results/all/?keywords=%23qualityengineering&origin=HASH_TAG_FROM_FEED #IndustrialAutomation https://www.linkedin.com/search/results/all/?keywords=%23industrialautomation&origin=HASH_TAG_FROM_FEED #LessonsLearned https://www.linkedin.com/search/results/all/?keywords=%23lessonslearned&origin=HASH_TAG_FROM_FEED

    linkedin.com

    www.linkedin.com/search/results/all/?keywords=%23pamilocell&origin=HASH_TAG_FROM_FEED

  • Lali Singh@lali-singh· 543 viewsin General Discussion

    Has anyone built the 3d Printed version of the Mk5 ? how much payload difference would it be with the 3d printed version vs the aluminum version ?

    Loading comments…

    Loading comments…
  • John Dammeyer@johndammeyer· 609 viewsin General Discussion

    A while ago (a long while) I wanted to build the XY mechanism to duplicate the desktop punch press as shown in this article.  https://www.cnccookbook.com/a-quick-way-to-make-lots-of-holes-with-cnc/

    Well 3D printing has taken over and the need for this or for that matter my 3:1 shear/brake/roller has almost vanished.

    The idea of using the AR3/4 hard to hold a piece of sheet metal and move it under the punch and then pause while an operator pulls on the handle is kind of a neat idea.  And that the Robot can read G-Code implies simple XY positioning along with the halt or pause.

    So suppose two limit switches are set up by the press.  One to detect the back edge and one to detect one side.  That becomes the HOME location for the XY geometry.  After that at hole at X25 Y25 should be easy right?

    Is there any sort of app note or discussion on how to do this?

  • John Dammeyer@johndammeyer· 795 viewsin Robot Builds

    The box is complete with slide out shelf for the power supply.  Not sure yet how I'll finish it.  Paint or maybe get fancy with oak trim.  Once the glue is dry it's time to finish sand it and add the cross grid at the 50mm spacing as suggested by Chris.

  • roland.petrasch@rolandpetrasch· 850 viewsin Robot Builds
    Edited

    🤖 𝗥𝗼𝗯𝗼𝘁𝗶𝗰 𝗠𝗮𝗻𝗶𝗽𝘂𝗹𝗮𝘁𝗼𝗿 𝗖𝗼𝗰𝗸𝗽𝗶𝘁: Easy install, setup, first tests for 𝗔𝗥𝟯/𝟰 with 𝗥𝗢𝗦𝟮, 𝗠𝗼𝘃𝗲𝗜𝘁𝟮, 𝗚𝗮𝘇𝗲𝗯𝗼, 𝗟𝗶𝗰𝗵𝘁𝗯𝗹𝗶𝗰𝗸, 𝗧𝗶𝗺𝗲𝘀𝗰𝗮𝗹𝗲𝗗𝗕 & more

    Simulate on a Mac/Linux, then run the same software on the real robot. That's the idea behind AR3 Robotic Arm Cockpit, which Richard and I are building for the open-source Annin Robotics AR3 arm.

    What works today (simulation):

    🦾 ROS 2 Jazzy with MoveIt 2 (OMPL and Pilz) and ros2_control

    🏭 A Gazebo Harmonic workcell with table, bins, objects and overhead and wrist cameras

    💻 Runs on Linux and Apple Silicon Macs (Docker, amd64 and arm64), with RViz, RViz in the browser or Lichtblick

    🖥️ AR3 Cockpit, a local web app for install and setup, starting and stopping stacks, monitoring and MoveIt scenarios. It shows the CLI command behind every action

    🗄️ Telemetry and run data captured in TimescaleDB

    💬 An assistant that answers from the project docs, with sources, running locally on Ollama or on Claude.

    Next steps: support of AR4, the real arm, multi-webcam perception, then pick, place and sort.

    On the research menu:

    • Data quality for robotics, high priority: checks and cleansing for sensor data, AI-ready data from simulation and physical-twin runs, and evaluation of AI results, using Grafana and other DQ tools

    • VLA fine-tuning (OpenVLA-OFT, π0 family) on cloud GPUs

    • Language task parsing with a local VLM on a MacBook

    • A learned world model with a sampling-based planner

    • Better perception with lidar and depth cameras

    • NVIDIA Isaac Sim and MuJoCo, as Ekumen shows for the AR4

    The code is on GitHub:

    Loading comments…

    Loading comments…