Tutorials
Step-by-step guides, by goal. Most give the goal, prerequisites, time, result and what to do next. The labels are defined in Status labels.
flowchart LR
A["collect<br/>teleop"] --> B["check<br/>check-dataset"]
B --> C["train<br/>lerobot-train"]
C --> D["run<br/>roboloom infer"]
D --> E["evaluate<br/>label"]
E --> F["collect with interventions<br/>policy_teleop"]
F --> B
First steps
-
Start, record and check a dataset with examples/mock.
-
Run a mock arm from your own follower class and template.
Rakuda2
The order is in Rakuda2 path.
-
Rakuda2: configuration and read check
Create the configuration, read the buses, and go once through with the torque off.
-
Measure zero points, travel and the gravity model, and tune the leader.
-
Rakuda2: teleoperated collection
Drive with the leader and record with the RealSense and DIGIT.
Policies
-
Rakuda2: from collection to inference
Check, train with LeRobot, run the policy and record the verdicts.
-
Rakuda2: evaluating with interventions
Take over from the policy with the leader; intervened rows are marked.
-
Train the sample chunk_mlp.py on a mock recording and run it with infer.
Robots in development
The order for PiPER + Pika is in PiPER + Pika path. Bringing up a robot follows the same stages on every robot.
-
PiPER + Pika: from robot setup to teleoperation
Identify the devices with
roboloom setupand teleoperate for the first time with a loose configuration. -
Record with mock Pika Senses through IK and run a LeRobot policy.
-
Hand-held collection with Pika Sense
Collect with Senses in the hand, no robot, and convert for training.