Fall 2026 GRASP SFI - Fernando Cladera, University of Pennsylvania

In this talk

1. 1:59

   Why disasters need autonomous robot teams

   Earthquakes and fires can destroy communications infrastructure and leave responders too busy to operate robots or interpret large volumes of robot data. Cladera emphasizes that rescue time matters: after the 2023 Turkey earthquake, most reported rescues occurred within 48 to 72 hours. Robots should give responders timely, actionable information, such as which casualties to reach first.

2. 8:01

   Sharing information across aerial and ground robots

   Cladera pairs UAVs, which can quickly survey large areas, with ground robots such as Spot, which can approach casualties and validate what aerial cameras detect. His team uses opportunistic communications: robots exchange stored messages when they come into range, with UAVs acting as both mappers and data relays. In a real-world mission presented at NASA’s 35th anniversary event, one UAV and three ground robots shared maps and explored a large area without continuous connectivity.

3. 21:36

   Compact semantic maps for changing missions

   To make missions more adaptable, the team combines FINE’s high-level task planning with zero-shot segmentation models such as Grounded SAM. Instead of transmitting dense maps, the UAV can share a scene graph containing selected objects and navigation points, reducing map size by an order of magnitude. Operators or robots can change the labels of interest during a mission, for example shifting attention to barrels and people after a chemical spill.

4. 27:33

   Event cameras for fast motion and difficult lighting

   Event cameras report brightness changes at individual pixels, which helps in uneven lighting and fast-moving scenes. Cladera’s team used one to predict where a fast ball would hit and direct a robot arm; the catch sequence took about 300 milliseconds, with roughly 80% success. For aerial mapping, fusing event-camera data with RGB images recovered sidewalk detail lost in bright conditions, though event cameras can be noisier.

5. 39:04

   Building toward long-term field autonomy

   Cladera describes long-term autonomy as operating for extended periods without human supervision, a key need when responders cannot continually manage robots. He points to small, collision-resilient UAVs, better ways to balance onboard compute with size, weight, and power limits, and processing event-camera data close to the sensor. The goal is to keep robots useful across changing field conditions and continue their work when operators are unavailable.