Trident AMR: Design and Simulation of a Differential-Drive Pick-and-Place Robot

Brief

Trident is a compact differential-drive robot designed and built for a WorldSkills competition task: follow a line, then pick and carry coloured coins to the corresponding bins of their color. The physical robot was designed in SolidWorks, 3D-printed in PLA, and wired within fifteen days. It won the district round of the competition. Following a sensor failure at the state round, I developed a high-fidelity Gazebo simulation and complete autonomy stack to prove the full mission end-to-end, verified and scored against ground truth by an automated referee node.

Problem

Building a competitive autonomous mobile robot under strict dimension and timeline limits poses sharp mechanical and perceptual challenges:

• Blind zone: The Intel RealSense D435 depth camera sits 85 mm above the floor and cannot see the floor within 0.15 m of the robot. Global navigation alone cannot position the robot accurately enough to grab small coins.
• Payload & tipping constraints: Lifting coins forward of the front support risks tipping the lightweight platform. Moreover, adding a vertical lift axis increases mass, wiring, and actuator complexity.
• Traction limits: Mass properties from the CAD model reveal that the ball caster carries 82% of the robot’s weight, leaving only 18% on the driven wheels. Acceleration and braking must be strictly capped to prevent wheel slip and loss of odometry.

Approach

• Caging Gripper Design: Replaced standard friction-grasp lifting with a horizontal sliding cage. Two parabolic jaw profiles ($y^2 = 4fx$) coupled via equal involute gears are driven by a single servo. Coins are corralled inward along the profile and slid along the floor without lifting. Jaw release ridges push the coin free upon opening.
• Dynamic & Stability Analysis: Computed exact mass properties from CAD. With static wheel loading at 1.16 N and caster loading at 5.12 N, drive acceleration is limited to 1.0 m/s² to stay well within traction limits (1.8 m/s²) and provide a 2× tipping safety margin.
• Sensor Fusion & Two-Stage Docking: Combined global SLAM with local tactile/reflectance guidance. Nav2 brings the robot to a station pose just short of each stem. An 8-channel infrared array (pitch 15 mm) then takes over with a PD line-following law whose damping is set by the array’s forward offset ($L = 48\text{ mm}$), stopping with sub-centimeter accuracy at crossbars.
• Simulation in Gazebo & Autonomy Stack: Exported URDF meshes directly from CAD. Integrated RTAB-Map for visual SLAM, Nav2 with Regulated Pure Pursuit for path tracking and dynamic obstacle inflation, and a custom mission state machine.
• Ground-Truth Referee: Built an independent evaluator node in ROS 2 that monitors true coin placements directly from simulator state, eliminating evaluation bias.

Results

• 100% Mission Success: Delivered all 36 coins across 12 full missions (spanning four obstacle layouts) with an average mission completion time of 300 s.
• Obstacle Avoidance: Successfully navigated around unmapped obstacle boxes with less than 5 s total time penalty compared to unobstructed runs.
• Precision Docking & Placement: Reached pick markers within 7 mm laterally and deposited coins within 21 mm of the center of target drop squares (which provide $\pm 125\text{ mm}$ tolerance).
• Line Following: Maintained line tracking within 6.0 mm RMS across noise levels up to 0.2 V.

Status / Next

The simulated mission is complete, fully tested, and repeatable. Next hardware steps include adjusting the caster and camera positions to shift center of mass rearward for greater wheel traction, eliminating breadboard wiring with a custom PCB, and validating the autonomous mission on the physical platform.