Repository navigation
UART baud rate characterization on a production robot (F31) #260
Replies: 2 comments
|
Addendum: the third link — RS485 SerialBus (core ↔ plexus), measured on real hardware. The original campaign covered UART0 (Jetson↔core) and UART1 (core↔P0 expander); the SerialBus numbers below close that gap. Setup: F31 with a real plexus (ESP32-S3) as bus peer, half-duplex RS485 with auto-direction transceivers (core UART2, RX 13/TX 4; peer Serial 39/40), both nodes on the
Key observations:
Two operational pitfalls surfaced: Caveat for transferring these numbers to other platforms (e.g. the drum robot): this was short-run bench-style wiring on F31. Cable harness and slip rings may cost margin, and 460,800 has zero headroom to the transceiver limit — 230,400 already carries the full 100 Hz payload and is the conservative operating point; 460,800 should be confirmed on the target wiring first. |
|
Addendum 2: RS485 SerialBus on the drum robot (W1/rb70) — production wiring, production firmware. Follow-up to the F31 addendum above: same dual-side checksum-counting methodology, but on the drum robot's built-in electronics (core ↔ plexus through the real drum cable harness) and stock Lizard v0.13.0 on both nodes instead of the
Key observations:
Conclusion for the drum robot: 460,800 is validated on the target wiring with demonstrated margin and is the recommended default. |
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Since #218 made the UART0 baud rate configurable, we measured how far both serial links can actually go on real hardware. Every Lizard line carries an XOR checksum, so corruption is directly countable without extra instrumentation.
UART0 (Jetson ↔ core), core→Jetson direction
Key observations:
UART1 (core ↔ P0 expander)
The surprise: the link is clean at every rate — the problem is payload, not baud.
Lengthening the P0's output line step by step gives a slope of 91 µs/byte (theoretical char time at 115,200: 86.8 µs/byte) — the P0's loop period is its transmit time. Working backwards, the production payload is ~255 B/cycle against ~115 B/10 ms of capacity: the TX buffer fills,
uart_write_bytesblocks, and the 10 ms loop stretches to 22 ms. Every P0-side signal (bumpers, battery status) updates at 45 Hz instead of 100 Hz on affected robots today.Suggested takeaways
Method & conditions (for reviewers)
core.set_baudrate()+ restart; rates > 921,600 via a build with an extended whitelist on the identical source base.p0.flash(true)(the replicator copies the core's NVS — see companion issue).read_line) so the expander handshake survives cross-baud boot chatter; they do not affect measured values.Resulting issues and PRs
p0.flash()silently replaces the expander's NVS with a copy of the core'sdisconnect()(3 s error spam)All reactions