

Teaching Drones to See with Margarita Chli
May 8, 2025
Margarita Chli, Head of the Vision for Robotics Lab at ETH Zurich and the University of Cyprus, shares her groundbreaking work in teaching drones to perceive their surroundings. She discusses the challenges of robotic vision and the fascinating applications in areas like wildfire management and search and rescue. The conversation also highlights the ethical implications of dual-use technologies and the importance of interdisciplinary collaboration. Margarita emphasizes how curiosity fuels scientific discovery and the evolving role of AI in robotics.
AI Snips
Chapters
Transcript
Episode notes
Vision Inspired By Humans, Built For Action
- Robotic vision is inspired by human vision but focuses on actionable scene understanding for autonomy.
- Machines can match or exceed human vision in specific tasks by tailoring sensors and algorithms.
Design Vision Around Hardware Limits
- Vision systems must be designed around the sensor and compute limits of the platform.
- Algorithms change dramatically if you have few pixels, low compute, or limited battery on a small drone.
Swarm Views Beat Single Perspectives
- Multiple robots viewing a scene can produce richer information than a single human viewpoint.
- Sharing views at runtime enables better decisions but raises hard questions about what to exchange.