During experiments in autonomous robotics, researchers collect a lot of data about what the robot sees, what it interprets, what it thinks and what it plans to do. In order to makes sense of these data and correctly interpret the results of the experiments, one often has to come up with clever ways to visualize these data. The resulting images and animations are often insightful, occasionally fun and maybe also beautiful (at least for the eyes of nerds). This talk goes through several such examples, covering recent research topics in autonomous robotics that took place at the Dalle Molle Institute for AI in Lugano. It will also present some geeky diversions that have no real use besides visual delight.

Devoxx

A professor at the Dalle Molle Institute for AI presents how data visualization is essential for understanding and debugging autonomous robotic systems. Drawing from over a decade of research, he walks through multiple projects: a drone following forest trails using early CNNs, nano-drones tracking people at 100fps, gesture-controlled drones, and a social robot offering chocolates by reading body language. The core message is that generating rich visualizations of what robots perceive and plan is critical for debugging, insight, and communication. The talk closes with personal physics simulations made for fun using Python, Matplotlib, PyBullet, and ROS.

Visualizing robot minds, and other fun stuff by Alessandro Giusti