Quick start

These procedures play the 100 mm circle demo. Read Safety first.

Before you start

• Clone the repository and install the Julia dependencies:

  julia --project=. -e 'import Pkg; Pkg.instantiate()'

• For the plots, make a Python environment in tools/python/ with pip install -r requirements.txt matplotlib numpy.
• Make sure that the space around the arm is clear.

Plan the circle

1. Run the planner:

   julia --project=. scripts/plan_circle.jl

2. Make sure that the output shows plan: 15.0 s and no assertion errors. The plan goes to tools/python/plans/circle.csv.

The plan starts and ends at the zero pose (arm upright).

Play the circle on the ATOM (WiFi)

The ATOM must run the controller firmware.

1. Connect the 12 V supply. The LED matrix shows blue, then green. Green means that the arm holds its pose.
2. Find the ATOM’s address in its status log (UDP broadcast, port 5005), or use the last known address 192.168.1.107.
3. Move the arm to the zero pose:

   import MyCobot
   link = MyCobot.AtomLink("192.168.1.107")
   MyCobot.atom_move_to(link, zeros(6); duration=6.0)
   close(link)

4. Play the circle:

   julia --project=. scripts/play_plan.jl tools/python/plans/circle.csv --atom=192.168.1.107

5. Make sure that the output shows result = "done" and late_cycles = 0.

Play the circle from the laptop (FT232)

The ATOM must be idle (the controller firmware is idle when it does not play a plan).

1. Connect the FT232R to the base: GND, and the base pins 13 and 14. See Robot and wiring.
2. Connect the 12 V supply.
3. Play the circle. The script sets the FT232R latency timer to 1 ms:

   julia --project=. scripts/play_plan.jl tools/python/plans/circle.csv

To move the arm to zero from the laptop, use MyCobot.move_to(sp, zeros(6)) on an open serial port.

Examine the result

1. Calculate the flange path from the recording:

   julia --project=. scripts/trace_recording.jl tools/python/recordings/<recording>.csv

2. Plot it:

   cd tools/python && python plot_circle.py recordings/<recording>_path.csv

Improve the result with learning control

1. Make the next plan from the last plan and its recording:

   julia --project=. scripts/ilc_step.jl tools/python/plans/circle.csv tools/python/recordings/<recording>.csv

2. Play the new plan (circle_ilc1.csv) as above.
3. Do the two steps again with the new plan and its recording. Three iterations reduce the error from 5 mm to about 1 mm RMS.