Record a Session
Every state sample and every command as JSON lines, and how to read the file back.
robot.record(path) writes every state sample the robot sent and every command the script
sent to one file, one JSON object per line, in order. Nothing is sampled or averaged.
record_and_replay.py records a few seconds of state, then reads the file back. It sends
nothing to the robot:
# Every state sample and every command sent while the block runs goes to PATH.
with Robot().connect() as robot, robot.record(PATH) as recording:
time.sleep(SECONDS)
print(f"recorded {recording.samples} states and {recording.commands_written} commands")
# Each line is one JSON object; "kind" says whether it is a state or a command.
lines = list(load(PATH))
states = [line for line in lines if line["kind"] == "state"]
print("robot modes seen:", dict(Counter(s["mode"] for s in states)))
if len(states) > 1:
rate = (len(states) - 1) / (states[-1]["t"] - states[0]["t"])
print(f"state rate {rate:.0f} Hz")record is a context manager: recording stops when the block exits, including when it exits
because of an exception, which is the run you want the file for. recording.samples and
recording.commands_written count what went in.
Read It Back
menlo.asimov.recording.load(path) yields one dict per line. kind is "state" or
"sent", t is time.monotonic() when the line was written (subtract the first line's t for
the time since the recording began), and a state line carries the
robot mode as mode beside the fields of State.
What It Is Good For
| Use | Why |
|---|---|
| A bug report | a fall or a robot that did not move is easier to read from the file than to describe |
| Comparing runs | the same script on two robots, or before and after a change, gives two files with the same shape |
| Checking a loop | the t of the sent lines shows whether a control loop held its rate |
Related
- Read State: the fields in a state line
- SDK Reference:
record,Recordingandload