Robotics / Autonomy

Tank-Track Mobile Platform

Joel Johnston 2026-04-06 Pre-stroke design

Tank-Track Mobile Platform

Author: Joel Johnston Date: 2026-04-06 Domain: Robotics / Autonomy Stroke Timeline: Pre-stroke

Abstract

Tracked robot with onboard compute. Dual DC gear motors driving rubber tracks with skid-steer differential drive, Raspberry Pi 4B as main compute, ESP32 as motor controller, ultrasonic sensors for obstacle avoidance, and MPU-6050 IMU (inertial measurement unit) for orientation tracking. Carbon fiber chassis with 3D-printed brackets from the parametric OpenSCAD library. $245 total build. roboNet mesh integration makes this a mobile network node. Future: SLAM (simultaneous localization and mapping) with RPLiDAR.

Hardware

Drive System

  • Motors: 2x DC gear motors, 12V, 200RPM, with quadrature encoders
  • Drive type: Skid-steer differential — no steering mechanism, turning is achieved by varying track speeds
  • Motor driver: L298N dual H-bridge, 2A per channel, PWM speed control
  • Tracks: Rubber tank tracks, 35mm wide
  • Power: 11.1V 3S LiPo 2200mAh (main), 5V 3A BEC (buck-boost for compute)

Compute Stack

  • Main SBC: Raspberry Pi 4B (4GB) — navigation, sensor fusion, mesh client, user interface
  • Motor controller MCU: ESP32 — receives velocity commands from Pi over USB serial, drives L298N, reads encoder feedback, runs velocity PID loop

This split keeps real-time motor control on the ESP32 (deterministic, loop time <1ms) while the Pi handles non-deterministic tasks (navigation, network, CV).

Sensors

  • Ultrasonic distance sensors: HC-SR04 (3x) — front, left-front 45°, right-front 45°
  • IMU: MPU-6050 (6-axis) — 3-axis accelerometer + 3-axis gyroscope, I2C
  • Encoder feedback: Quadrature encoders on both motors (via ESP32)

Chassis

  • Plates: Carbon fiber 3mm (custom cut, 2x — top and bottom)
  • Brackets: 3D-printed PETG from parametric OpenSCAD library
  • Standoffs: M3 aluminum, configurable height for different payload heights
  • Payload capacity: approximately 500g on top plate

Software Architecture

ESP32 Motor Controller

Firmware responsibilities:

  • Receive {left_vel, right_vel} commands over USB serial (JSON or simple binary protocol)
  • Run velocity PID loops independently for each track (encoder feedback)
  • Report actual velocities and encoder counts back to Pi at 50Hz
  • Handle watchdog — if no command received in 500ms, stop motors (failsafe)

The ESP32 firmware is deterministic. Command-to-motor-response latency: <5ms. The Pi does not need to worry about real-time constraints.

Raspberry Pi Navigation Stack

ultrasonic sensors → obstacle map (2D grid, robot-centered)
                           ↓
                    path planner (reactive avoidance)
                           ↓
                    velocity commands → ESP32 → motors
                           ↑
                    IMU → heading hold (Kalman filter)
                    encoder odometry → position estimate

Obstacle avoidance: VFH (Vector Field Histogram) — simple reactive planner. Scans ultrasonic readings, computes histogram of obstacle density by angle, finds clearest direction, steers toward it.

Heading hold: When no steering correction is needed, the Pi uses IMU gyroscope integration to maintain the commanded heading. Encoder odometry provides secondary reference. Simple complementary filter (not full Kalman — sufficient for indoor use).

roboNet Mesh Client

Pi runs the standard roboNet mesh client. The robot registers as a node with capabilities:

{
  "node_type": "mobile",
  "capabilities": ["drive", "ultrasonic_scan", "imu", "odometry"],
  "location": "dynamic"
}

The mesh can dispatch tasks to the robot:

  • Drive to coordinates (relative to home)
  • Scan an area (systematic coverage pattern)
  • Report sensor data on demand

Full BOM

Component Cost
Raspberry Pi 4B (4GB) $55
ESP32 development board $8
DC gear motors with encoders (2x) $30
L298N motor driver $8
Rubber tank tracks (pair) $25
3S LiPo 2200mAh $22
5V BEC (buck converter) $8
HC-SR04 ultrasonic sensors (3x) $9
MPU-6050 IMU $5
Carbon fiber plates (2x) $40
3D-printed brackets (PETG) $5
Hardware, wiring, connectors $15
Total ~$245

Autonomy Stack — Current and Future

Current: Reactive Navigation

The current implementation uses reactive navigation — no map building, no global planning. The robot responds to immediate sensor readings. Effective for obstacle avoidance in dynamic environments. Insufficient for destination navigation.

Future: SLAM with RPLiDAR

Adding RPLiDAR A1 (2D LiDAR, 360-degree, 12m range) enables:

  • Mapping: build an occupancy grid of the environment while driving
  • Localization: use the map to determine position (particle filter or scan matching)
  • Path planning: A* or Dijkstra on the occupancy grid for destination navigation

Software stack: slam_toolbox (ROS2) or a custom lightweight SLAM implementation. The Pi 4B has sufficient compute for 2D SLAM at normal robot speeds.

Future: Multi-Robot Coordination

Through the roboNet mesh, multiple tank-track platforms can share maps and coordinate:

  • One robot maps, others receive the map
  • Distributed area coverage (no overlap)
  • Rendezvous for payload transfer

The coordination is handled by the mesh task dispatch system — no robot-specific coordination code. Same protocol as any other distributed task.