Calibrate
Calibration writes four facts into conf/modules/right_arm.yaml, per joint: how far the encoder travels (real_range_min, real_range_max), where zero sits (center_offset), and whether the sim joint turns the same way as the metal (drive_mode). The gripper gets the same treatment, plus a closed pose.
The long version, with the status-board screenshots, is Calibration. This page is the path through it for the arm you just built. Do it on the Pi, which is the machine the adapter is plugged into. Stop run-robot first. The bus can be open once.
solarian --deployment lab calibrate
Pick right_arm when it asks.
What you will be asked
The command uses the same three steps as the reference page.
Joint limits. Move every hard-stop joint through its full travel by hand, with torque off, while the status board shows MIN / POS / MAX. Wrist roll on this arm is continuous: it is printed and skipped, because it has no stop. If a joint's travel crosses the encoder seam at 4095→0, the command stops and tells you to refit that horn at mid-travel. Do it. No offset fixes a seam. The servo computes "goal minus present" without wrapping, so a goal on the other side of the seam drives into the stop.
Home. The sim window opens at the model's zero pose, which is not a live copy of your arm. Move the real arm until it matches that picture. That pose is mid-travel for the joints that have stops. Leave a few degrees of room each way.
Direction. The sim starts following the real arm. A red arrow marks one hinge at a time. Move that joint. If the sim goes the other way, answer N and the command flips drive_mode.
Answer ENTER / Y / N in the sim window. The terminal accepts them too.
When it finishes, look at conf/modules/right_arm.yaml. Each joint should now have real_range_min, real_range_max, center_offset, and drive_mode. Those numbers belong to your arm. Someone else's YAML will not. A home: block is the normalized pose you matched. It is not the home pose Tutorial 4's move skill will use. Same word, different file.
The gripper
Run solarian --deployment lab calibrate again and pick right_gripper. Grippers stop after the limit sweep plus one physical closed pose: fingers together, then confirm. The range lands on gripper in right_gripper.yaml.
Fingertip spacing is not linear in encoder counts. After the arm moves well, solarian calibrate-gripper-ranges records a table from a ruler so that "40" means 40 millimetres. You can teleop without that table. A policy that opens to a width in millimetres should not. The command's prompts are the procedure; the reference is the gripper section of Calibration.
solarian calibrate-arm and solarian calibrate-gripper are a different measurement. They time the loop (speed and latency) and write max_speed and action_latency. Skip them until a skill is moving the arm in Cartesian space. Joint teleop does not use those numbers.
Next: drive the real arm.