How ESP32–SBC communication became smaller, faster and safer.

Initial state: one communication path for different jobs

Telemetry, control commands and larger spatial information shared assumptions about frequency and reliability. A telemetry frame was about 75 bytes, a control frame about 20 bytes, and the SBC decision loop ran at 30 Hz — one cycle every 33.3 ms.

Diagnosis: reliability and frequency are not the same requirement

Fresh telemetry loses value quickly, so a late packet is often worse than a dropped one. A drive command is different: ordering and acknowledgement matter. Spatial data is heavier and changes more slowly. Treating all three streams identically created avoidable overhead and coupled independent failure modes.

No backlog: telemetry should deliver the newest state, not the history of a queue

In a real-time path, FIFO can preserve samples that are technically valid but already obsolete. If the receiver stalls briefly, replaying t−3, t−2 and t−1 only delays arrival at state t. An older telemetry sample may therefore be skipped so the receiver can operate on the newest valid frame.

Before: queue and replay sample → serialize → queue → transport → queue → parse → decision

→

After: latest-state / no backlog sample → current frame → transport → validate → current state

Telemetry Latest value, sequence number and data age. Stale samples are not replayed.

Control Ordering, ACK and bounded retry. Commands are not treated like disposable telemetry.

Map / spatial data A separate lower-rate stream prevents heavier transfers from blocking fresh state.

Decision: separate the channels

Fixed frames and bit packing: the parser reads offsets instead of interpreting message descriptions

The 75 → 28 B and 20 → 9 B reductions are not merely compression. The representation changed to a fixed binary frame with known-width fields and bit-packed flags. The receiver reads predictable offsets and validates version, sequence, timestamp and checksum.

7

6

5

4

3

2

1

0

Bit flags READY, ERROR, LIMIT, ESTOP and SENSOR_OK can use individual bits instead of separate textual fields.

Fixed offsets Marker, version, sequence, timestamp, flags, payload and checksum occupy predictable positions.

Less work Fewer bytes, copies, serialization steps and parsing work on the critical path.

Important: no-queue semantics apply to fresh telemetry. Control still keeps ordering, ACK and bounded retry.

Why the watchdog stays on the ESP32

The Linux computer can restart, stall or lose its link. The emergency reaction therefore cannot depend on the same path that may have failed. ESP32 tracks the age of the last accepted command and stops outputs locally after the configured timeout.

if (now_ms - last_valid_control_ms > CONTROL_TIMEOUT_MS) {
    drive_left = 0;
    drive_right = 0;
    state = SAFE_STOP;
}

Result and interpretation

Smaller frames reduced serialization and parsing work, while independent channels prevented large or reliable transfers from blocking fresh telemetry. In the measured system, the decision loop increased from 30 Hz to 60 Hz. The key improvement was architectural: each data class received the transport and failure policy it actually needed.