Efficiency in Logistics: Combining Data Science & AI

Michael Nederhoed and three classmates worked with a robotics logistics company on automating factory operations. The team built a system that lets robots identify objects by image recognition, calculate 3D coordinates, drive seven motors and carry out pick-and-place operations - replacing the barcode and pressure-sensor methods the company had been using.

The solution combined several AI systems, working entirely from algorithmic and image-based approaches. Michael contributed data analysis, Python programming and AI expertise; the team rotated scrum master duties every two weeks.

Reinforcement learning

The most interesting part was watching different AI techniques work together. The team used reinforcement learning to train the robot arm’s movement between points:

“By rewarding the robot when it performed correctly and penalizing it for errors, we created a reward function”

• one that encouraged the behaviour they wanted and discouraged unsafe actions such as applying excessive force.

Afterwards

Michael wants to build AI models and solutions at the intersection of data science, AI development and consultancy, rather than making dashboards.