Autonomous robot prototypes.

At Analytical Robotics, I helped develop working prototypes for a multi-robot system. The work covered the full software stack for autonomy—localization, perception, planning, and control—alongside the electronics and mechanical iteration needed to make those systems useful on real hardware.

Prototype work is integration work. A clean algorithm is only valuable when encoders, sensors, power systems, motor drivers, mechanical tolerances, and communications all agree long enough to complete the task.

Version one: learn quickly

The first platform proved the basic architecture. Stepper motors, time-of-flight sensors, and Decawave UWB boards provided the foundations for movement, proximity sensing, and localization. Its open frame kept every system reachable while the design was still changing.

Version two: carry the lessons forward

The second generation moved to a larger, purpose-cut aluminium frame and four 250W DC motors. It offered more space, power, and robustness—but also made cable routing, serviceability, component placement, and weight much more consequential.

Breadboards are excellent while the question is “does this component work?” They are much less useful once the question becomes “can another person diagnose this machine?” The electronics evolved from exploratory wiring toward clearer groupings, stronger connections, and components positioned for access.

The system brought together motor drivers, an RS-485 interface for AMT shaft encoders, a four-channel DAC, a Teensy 4.1, and an NVIDIA Jetson Nano. That mix made the robot a useful test bed for both low-level control and higher-level autonomy.

• Keep experimental systems accessible until the architecture is genuinely stable.
• Clean wiring is not cosmetic; it shortens every debugging session that follows.
• Mechanical, electrical, and software decisions must be reviewed as one system.

A single robot moving is a milestone. Multiple robots sharing a task exposes a different class of problems: relative position, timing, repeatability, collision avoidance, load handling, and recovery when one unit behaves differently from the rest.

The result was not a showroom object; it was a working engineering platform. It made autonomy visible, measurable, and improvable, while giving the team a concrete system for customer conversations and the next round of technical decisions.

The enduring lesson was that prototype quality is not polish for its own sake. Good structure creates faster experiments, more reliable evidence, and a shorter route to the next correct decision.