Personal project
Started
As of
Build log · Navio2 · ArduRover
Wheels off the ground.
A Traxxas TRX-4 Ford Bronco, orange, with an Emlid Navio2 autopilot on a Raspberry Pi in place of its receiver. ArduRover drives the steering servo, the ESC and the crawler's aux servos over ordinary hobby PWM; a RadioLink handset stays bound as the override; a ground station watches over Wi-Fi. Drive-by-wire, an arm switch, battery telemetry and a lidar are verified on the bench. A forward/reverse ESC has just gone in, still to be calibrated, and GPS missions are next.
Every step so far has happened on a stand, with the truck unable to drive off the bench. This page is the plan in five phases, what is verified against it, the wiring as built, the parameters that mattered, and the mistakes that cost time. It moves as the build does.
Five phases
Phase 0
In progress
Prove the autopilot can command the truck and the operator can always take it back.
Verified
5 of 7
Phase 1
In progress
Sensors online, operator still in charge.
Verified
1 of 6
Phase 2
Planned
It drives a route on its own, inside a fence.
Verified
0 of 5
Phase 3
In progress
The truck reacts to the world, not only to coordinates.
Verified
0 of 6
Phase 4
Planned
Research-grade autonomy on a ROS 2 layer over the autopilot.
Verified
0 of 5
Each phase is only safe once the one before it is solid. verified in progress planned
In this order
Meter its receiver-lead BEC before it touches the board, and pull the red pin if it reads above 5.3 V. Calibrate its throttle range to the Navio2's 1500 neutral and 1100 to 1900 ends, wheels up, then check direction. It also took the winch's power with the old ESC's accessory plug, so the spare 5 A UBEC goes on a splitter leg for the winch.
The receiver lead and the lidar's USB plug both fell off within two days. A zip tie anchoring each cable near its plug, then the lidar goes back on and PRX_TYPE returns to 5.
Compare the reported pack voltage with a meter and scale the multiplier until they agree within 0.1 V, set the capacity to 5,000 mAh, then the low-voltage failsafe: about 3.5 V per cell, Hold or return to launch.
Then heading hold, the slower and tighter auto tuning, and the first waypoint mission on open ground.
The Module 3 cannot run on the autopilot's Pi. A companion computer carries it properly; a USB webcam on the existing Pi gets a picture off the truck sooner. The two are not exclusive.
Hall sensors and magnets on a driveshaft, counted by the Arduino: the Navio2's GPIO cannot count pulses reliably. Encoders are what make avoidance work indoors, and what makes "forward 1.5 m and stop" repeatable.
Most recent first
The XL-5 HV comes out for an ESC with forward and reverse, which should let the autopilot back up on its own; its BEC gets metered and its throttle calibrated before it drives anything, and the winch needs a new supply. An XT60 master switch goes on the pack lead, and the truck stays off between sessions, which turns out to be what every "dropout" was. The lidar's USB plug had come loose and blocked arming, so the lidar is switched out until it is strain-relieved.
The Camera Module 3 is ruled out on the autopilot's Pi: Emlid's image carries the legacy camera stack and the kernel the Navio2 needs, and neither knows the sensor. An RPLIDAR A1M8 goes on the Pi's USB instead: healthy, and silent to ArduPilot until the motor's DTR gating was found and fixed; then eight sectors of range flow and avoidance is enabled at half a metre. Earlier the same night a dead motor was traced to a throttle trim of 1100, and the receiver, off since the failsafe, was back on the link by the time avoidance went on.
CH9 becomes the arm switch and stick arming is turned off; only fully high arms, everything else disarms. The power module goes on the POWER port and the battery monitor reads the pack. And the receiver lead came off mid-session: the failsafe put the truck in Hold on its own, an unplanned test, passed.
The plan, the wiring and the bench results, written up here from the build documents.
Steering, throttle, reverse, the 2-speed, both lockers and the link-loss failsafe all answer the autopilot with the wheels off the ground. The output map is cleaned up: lockers on passthrough, a stray SERVO4 function set back to 0. Open items: a dead 3S pack and charger, and the power module.
Early in the month, wired backwards. It let the smoke out. The Navio2 and the Pi survived because the six-pin cable was not in the board at the time. Until the replacement, the 5 V servo BEC carried the Pi through the board's diode-OR, and there was no battery telemetry.
A stock TRX-4 Bronco, a Navio2 on a Raspberry Pi 3B, a RadioLink handset and an inventory of BECs, splitters and a power module. The plan is written in five phases and the first rule is set: nothing above 5.3 V touches the board.
A month means the day was not written down. Entries are added as the build moves.
As built
Principle
Everything downstream of the autopilot is ordinary hobby PWM, so the mechanical side is a drop-in: the Navio2's outputs plug in where the receiver's did. The care goes into two places, power and the override path. A real radio link stays bound to the autopilot so a switch on the handset hands the truck back at once, and nothing above 5.3 V is allowed to reach the board.
| From | To | Carries |
|---|---|---|
| Navio2 CH1 | Steering servo | Servo PWM at 50 Hz, 1500 µs centred3.3 V logic drives the servo directly. |
| Navio2 CH3 | ESC signal | PWM throttle, 1500 µs neutral, 1100 to 1900 µs endsSignal and ground only unless the ESC's BEC measures 5.3 V or less. The XL-5 HV's 6 V red wire came out; the new ESC's is to be metered. |
| Navio2 CH6, CH7, CH8 | Front locker, rear locker, 2-speed servos | PWM on RC passthroughEach channel mirrors its handset switch. |
| R12DSM SBUS lead | Navio2 PPM/SB column | All 12 channels on one three-wire leadThe board powers the receiver. Signal on the top row, ground on the board edge. |
| Power module, six-pin cable | Navio2 POWER port | 5.3 V, plus pack voltage and current on ADC pins 2 and 3Live since 2026-09-23. |
| Servo BEC, about 5.3 V | Navio2 servo rail | Servo power, and the Pi through the diode-ORBehind a 7.5 A fuse on the splitter. Never 6 V: the rail feeds the Pi. |
| LiPo pack | XT60 master switch on the pack lead, then the power module and the splitter | Pack voltageLeg 1 to the ESC, unfused; leg 2 to the servo BEC; a leg for the winch's UBEC next. The splitter's negative bus is the star ground. |
| RPLIDAR A1M8 | Pi USB | Serial over USB: ArduPilot SERIAL4, Lidar360 at 115200The motor is gated by the port's DTR line; a small service clears it after ArduRover starts. Unplugged since 2026-09-25, PRX_TYPE at 0, until it is back on. |
| Pi Wi-Fi | Ground station | MAVLink over UDPTo Mission Planner on the bench network. |
| VL53L1X | Navio2 I2C | Range, natively supportedPlanned. The barometer shares the bus at another address. |
| Pi Camera Module 3 | A companion computer's CSI port | Image data over the ribbon cableRecommended, not decided. Not the autopilot's Pi: its image cannot see the sensor. |
| Arduino TX/RX | Pi UART | Serial telemetryPlanned. The Arduino's 5 V TX needs a level shift into the Pi's 3.3 V RX. |
| Channel | Drives | On the handset |
|---|---|---|
| CH1 | Steering | Right stick, left and right |
| CH2 | Throttle | Right stick, up and downRCMAP_THROTTLE = 2 puts throttle on the same stick as steering; RC2_REVERSED sets which way is forward. |
| CH5 | Mode: Manual / Hold / Auto | Three-position switchThe override. Flip it and the truck is yours. |
| CH6 | Front diff lock | Toggle |
| CH7 | Rear diff lock | Toggle |
| CH8 | 2-speed transmission | Toggle |
| CH9 | Arm / disarm | Three-position switchOnly fully high arms. Middle and low both disarm. |
| CH10 | Light modes | TogglePlanned. |
| CH11, CH12 | Camera gimbal pan and tilt | KnobsPlanned, with the gimbal. |
ArduRover
Output map
An Ackermann rover needs two outputs assigned and every aux servo told to follow its switch. These are the values that made the truck drive, after the ones that did not.
| Parameter | Value | Why |
|---|---|---|
| SERVO1_FUNCTION | 26 | GroundSteering on output 1. |
| SERVO3_FUNCTION | 70 | Throttle on output 3. |
| SERVO3_TRIM | 1500 | Throttle neutral. At 1100 the ESC sees a permanent brake and the motor is dead. |
| SERVO6_FUNCTION | 56 | RCIN6: the front locker follows its switch. |
| SERVO7_FUNCTION | 57 | RCIN7: the rear locker follows its switch. |
| SERVO8_FUNCTION | 58 | RCIN8: the 2-speed follows its switch. Left at 0 the output sends no pulse, the servo never moves, and the transmission stays in neutral. |
| SERVO4_FUNCTION | 0 | Was a stray 102 from the shuffle. Unused outputs stay at 0. |
| RCMAP_THROTTLE | 2 | Throttle reads the right stick's vertical axis. |
| RC2_REVERSED | 1 | Flipped from 0 to 1 on 2026-09-25 to swap the stick's direction; not yet verified, and the new ESC may want it the other way. A radio calibration resets it. |
| MODE_CH | 5 | The three-position switch selects Manual, Hold or Auto. |
| RC9_OPTION | 41 | Arm/disarm on CH9. 41 is the value on Rover 4.0; it became 153 in 4.2. |
| ARMING_RUDDER | 0 | No stick arming. Steering and throttle share a stick, so a hard right at neutral used to arm the truck. |
| FS_ACTION | 2 | Hold when the radio link is lost. |
| BATT_MONITOR | 4 | Voltage and current from the power module. It was 0, which is why Mission Planner saw nothing. |
| BATT_VOLT_PIN, BATT_CURR_PIN | 2, 3 | The Navio2's ADC channels for the module's sense lines. |
| BATT_VOLT_MULT | 10.1 | Seed value. To be scaled until the reading matches a meter at the pack. |
| BATT_AMP_PERVLT | 17 | Seed value. Only affects the mAh count, not the voltage failsafe. |
| SERIAL4_PROTOCOL, SERIAL4_BAUD | 11, 115 | Lidar360 at 115200 on the USB serial port the lidar enumerates as. |
| PRX_TYPE | 5, now 0 | 5 is the RPLidar. Set to 0 on 2026-09-25 so a loose plug cannot block arming; back to 5 when it is remounted. |
| PRX_ORIENT | 0 | Upright, with the driver assuming the cable points to the rear; the mount has not been checked against that yet. |
| AVOID_ENABLE | 7 | Use the proximity sensor and fences. |
| AVOID_MARGIN | 0.5 | Half a metre. The 2 m default is for a full-size vehicle. |
| AVOID_BEHAVE | 1 | Stop, rather than slide along the obstacle. |
| OA_TYPE | 0 | No path planning yet; that comes with waypoint missions. |
| SCHED_LOOP_RATE | 50 | Left at the default. The docs suggest 400 for this lidar, and that is a lot to ask of a Pi 3 already flying the autopilot. |
Cheaper to read
Output map
The SERVOn_FUNCTION table got shuffled, and every symptom pointed somewhere else. The fix for this Ackermann truck is short: SERVO1 = 26, SERVO3 = 70, every aux servo on RC passthrough, everything else 0. A channel left disabled sends no pulse at all, so its servo never moves: that is how a 2-speed ends up stuck in neutral with the motor spinning and the wheels still.
Trim
After a round of settings changes the motor would not run. SERVO3_TRIM had moved from 1500 to 1100, so the ESC was being fed a permanent brake instead of neutral and refused to arm. Save a parameter file from a known-good truck; then a scramble is a diff, not a hunt.
Modes
Hold and Auto keep the throttle at neutral until there is a mission and a position, which from the bench looks exactly like a dead ESC. Manual is the mode for bring-up; Auto waits for GPS.
Arming
With steering and throttle on one stick, ArduPilot's stick arming meant full right at neutral throttle armed it. Now a dedicated switch is the only way: CH9 fully high arms, and middle or low disarms, so a knocked switch or a failsafe parking the channel fails toward safe.
Connectors
The receiver's three-wire lead came off mid-session and the failsafe caught it. Then the lidar's USB-C came loose and the pre-arm check reported bad lidar health. On a truck that vibrates and is handled constantly, every connector wants a zip tie anchoring the cable near its plug.
Reverse
The XL-5 HV reversed only after a double-tap: reverse, neutral, reverse. ArduRover holds a steady value, so in Auto it could brake and never reverse, and missions were planned forward-only. The forward/reverse ESC fitted on 2026-09-25 should remove the lockout; calibration will tell.
Power module
It let the smoke out. The Navio2 and the Pi survived because the six-pin cable was not in the board at the time. The replacement was wired correctly, and then read nothing in Mission Planner for a different reason: the battery monitor was switched off. Connecting the module is half the job.
Cells
A 4S pack tripped its over-voltage alarm, red and green, and it refused to arm. Its low-voltage cutoff was set to LiPo by holding EZ-Set for about ten seconds until the LED was solid green. The packs are 3S, 5,000 mAh.
Rail voltage
Its power port, servo rail and the Pi's USB are diode-ORed into one supply, so a 6 V BEC on the servo rail feeds 6 V to the Pi. The XL-5 HV's own 6 V receiver BEC was exactly that, which is why its red wire came out of the lead. The new ESC's BEC gets a meter before it gets a socket.
Camera
The Camera Module 3's sensor is known only to libcamera and a far newer kernel. Emlid's image is the older stack on a custom kernel, and that kernel is what carries the Navio2's driver. Upgrading the OS to chase the camera risks the autopilot, so the camera goes to a companion computer instead.
Lidar
Over the USB adapter the RPLIDAR's motor is gated by the serial DTR line. ArduPilot's driver expects direct-UART wiring where the motor always spins, and opening the port asserted DTR and switched it off, so the sensor sat silent with every parameter correct. A small service clears DTR a few seconds after ArduRover starts, and the range data flows.
Avoidance
Manual is pure stick passthrough by design. Simple avoidance acts in Acro, Steering, Loiter and Guided, and those need a velocity estimate the truck does not have indoors: no GPS fix, no encoders. So the lidar can see the room and the truck cannot yet act on it.
GPIO
ArduPilot's wheel-encoder driver needs GPIO interrupts, which Linux autopilots do not reliably provide, so the encoders go on the Arduino. I2C is fine on the board: the time-of-flight sensors plug straight into the Navio2 and ArduPilot reads them itself.
Seating
If Emlid's test reports the IMUs, barometer and GPS failing while the RCIO passes, the Navio2 is not fully on the header: SPI to the RCIO survives a partial seat and the sensor rail does not. Reseat and screw it down. An empty test while ArduRover is running is a different thing: the autopilot holds the sensors, so stop it first.
Switch
The Pi kept vanishing from the network mid-session, and a Wi-Fi power-saving fix went in to stop it. The Wi-Fi was probably fine: the truck was being switched off at the new XT60 master switch between sessions. Before hunting a network fault, ask whether the rover is on.
Inventory
| Part | Does | State |
|---|---|---|
| Traxxas TRX-4 Ford Bronco (92076-4) | Chassis: portal axles, two-speed transmission, remote-locking differentials | On the truck |
| Titan 550 brushed motor | Drive | On the truck |
| Forward/reverse ESC | Motor control, forward and reverse; expected to have no reverse lockoutFitted 2026-09-25. Make and model to be recorded; BEC to be metered; throttle range and direction to be calibrated. | On the truck |
| Traxxas XL-5 HV ESC | Motor control, 3S maximum, reverse only after a double-tapReplaced 2026-09-25. Its accessory plug had powered the winch. | Retired |
| Emlid Navio2 on a Raspberry Pi 3B | Autopilot: ArduRover 4.0 on Emlid's ArduPilot imageDual IMU, barometer, u-blox M8N GNSS, and an RCIO co-processor for 14 PWM outputs. | On the truck |
| RadioLink AT10II with an R12DSM receiver | Manual override, 12 channels over SBUSOne three-wire lead into the Navio2; the board powers the receiver. The lead has already fallen off once. | On the truck |
| Power module | 5.3 V to the Navio2 plus battery voltage and currentOn the POWER port since 2026-09-23; the monitor reads the pack. Multiplier still to calibrate. | On the truck |
| Emlid power module (the first) | The same jobWired backwards, early September. | Dead |
| 3S LiPo packs, 5,000 mAh | Main packNominal 11.1 V; the monitor reads it as 3S, as the XL-5 HV did. | On the truck |
| XT60 master switch | Cuts the pack from everythingOn the pack lead. The truck is off between sessions. | On the truck |
| XT60 1-to-4 splitter | Distribution block and star ground | On the truck |
| Servo BEC, 8 A, about 5.3 V at the rail | Servo rail supplyAlso carried the Pi through the Navio2's diode-OR while the power module was out. | On the truck |
| UBEC, 5 A | Accessory supply for the winchNeeded again: the winch lost the XL-5 HV's accessory plug. Proposed for a splitter leg at 5 to 6 V. | Spare |
| Traxxas Pro Scale lights | LightingRan from the XL-5 HV's 6 V receiver BEC; without a supply since the swap. | On the truck |
| Traxxas 8856 winch with the 8857 fob | RecoveryIts own 2.4 GHz link, no autopilot channel. Needs a supply since the ESC swap. | On the truck |
| Slamtec RPLIDAR A1M8 | 360° range for proximity, and later mappingGave ArduRover eight sectors of range over the Pi's USB on 2026-09-24. Its plug came loose; unplugged, with PRX_TYPE at 0, until it is strain-relieved and back on. | In hand |
| Raspberry Pi Camera Module 3, 12 MP | Vision: a stream first, then path following and marker dockingCannot run on the autopilot's Pi. Recommended for a companion computer; not yet decided. | In hand |
| VL53L1X time-of-flight sensors | Close-range stop, about 4 mI2C into the Navio2, which ArduPilot supports natively. One is easy; several need their addresses changed. | In hand |
| Hall-effect sensors | Wheel odometryOn a driveshaft, counted by the Arduino. | In hand |
| Magnets for the encoders | Wheel odometryOn order as of 2026-09-24. | Planned |
| SMT172 temperature sensor | Motor or ESC temperatureNot the logic-level shifter its part number suggests. Reads through the Arduino. | In hand |
| Companion computer | ROS 2, SLAM, the camera and the lidar's full scanA Raspberry Pi 5, 8 GB, is the candidate, with its own 5 V BEC of 6 A or more. | Planned |
| Arduino with a level shifter | Real-time peripherals over UARTEncoders, the temperature sensor, lights sequencing, a gimbal. | Planned |
| GNSS mast | Gets the M8N antenna away from the motor and ESC leads | Planned |
| Vibration-damped mounting plate | Isolates the IMU from the crawler's pounding | Planned |
Written from the build's own documents and bench notes. Nothing about the bench network is here on purpose.