Page 2: Messengers are sent if necessary

Contents

The emergency services should always be able to "fly the data required for AI evaluation themselves", demands Monika Gähler, Head of the Center for Satellite-Based Crisis Information at the German Aerospace Center (DLR). The German Federal Agency for Technical Relief (THW) is currently setting up a drone squadron for this purpose. This will make it possible to carry out initial analyses directly "in the car on site" shortly after the launch of an unmanned flying object. However, a "pure image flight with an uncoordinated video recording" is useless. After all, the data supplied must also be standardized in such a way that it can be evaluated quickly.

DLR is not starting from scratch when it comes to big data analysis, Gähler assures us: automated processes were already being used before the current AI hype. The center also makes it clear that the results generated with AI are experimental. During the earthquake relief in Turkey and Syria in February 2023, it was discovered that the Ahr valley experience could be used in terms of systems technology, but that the corresponding models still needed to be specifically trained. The technology does not yet offer "incredible accuracy". In the next three to four years, however, Gähler expects a major technical leap forward. However, everyone involved must remain prepared for failures and avoid dependencies as far as possible: "We have also printed maps again in the Ahr valley and sent them into the field with messengers." In crises, it is necessary to "know the key people".

Until more specific models are available, the DRK district association for the Aachen city region is making do with ChatGPT, thus fulfilling bureaucratic requirements, for example. A simple prompt issues the instruction to the bot: "Analyze image 1 and create a situation report using image 2 as a template," says CEO Axel Fielen, giving an example.

"Dating app" for the placement of helpers

According to Fielen, AI can already help to coordinate "independent" volunteers in the event of a disaster. In the Ahr valley case, the association was "overwhelmed by the number of people who came forward". At the time, eight colleagues were put on a hotline and asked to help via social media. Over the course of three days, 3,000 volunteers came together, which the employees entered into an Excel list. However, merging them with those seeking help then put the entire administration out of action for two weeks.

To eliminate the bottleneck, the DRK team has now developed an "app like a professional dating platform", says the boss happily. In the background, an AI evaluates the offers of help and matches them to the demand. Based on this matching, the operations management team can then find out which volunteers are potentially available. According to Fielen, a chatbot is also good at supporting direct communication between emergency responders and those affected in the event of a crisis by providing alerts, delayed feedback, help for self-help or "simple psychosocial support". A special version of the DRK Aachen is expected to be ready in the next three months.

AI for rescuing drowning victims

The Bavarian Red Cross (BRK) has taken a completely different direction. It wants to develop sonar devices with AI to help locate people, in particular to prevent drowning. Until now, it has been difficult to estimate distances on the water and it is often difficult to see underwater, says Benedikt Schlereth-Groh from the Nuremberg District Water Rescue Service, which is part of the BRK. Divers therefore have to feel their way for quite a long time. As a technical aid, sonar is already a method for locating and analyzing structures underwater using ultrasonic waves at 455 or 800 kHz. However, the resulting images are "not as high-resolution" and are noisy. This makes it difficult to find a missing person.

To solve this problem, the water rescue service therefore relies on the AI-supported Aqua-Eye underwater scanner. With this portable device, you have to go into the water, bring it under the surface and slowly rotate around its own axis in 30 seconds, says Schlereth-Groh, giving an insight into the process. A distance of up to 50 meters is then scanned. The AI then provides suggestions with circles if it is unsure, or with an X for a body that is likely to be discovered. In the latter case, the divers would be deployed directly.

The handheld could be used by numerous local groups and a lifeguard could use it to "quickly scan three ponds", explains Schlereth-Groh. In the future, the water rescue service also wants to test the Sonobot 5, a renamed mini-ship that travels across the water and automatically scans a larger area. However, at up to 80,000 euros, the price for this is significantly higher than that of the Aqua-Eye. The Nuremberg-based group has its own research project for sonar devices with AI support, running in parallel with the local technical university.

Machine people for "hostile environments"

At the provisional end of the chain, Straube from DFKI sees robotic systems for on-site helpers that establish physical interaction with the environment. The AI research center is working together with the Fraunhofer Institute of Optronics, System Technologies and Image Exploitation (IOSB) on the Robdekon project for disaster control to decontaminate "hostile environments". Such robots - including excavator systems - must be automated to such an extent that they "can operate without human intervention".

According to Straube, the DFKI is working with the THW to produce machines that "lift heavy loads". The EU project Deepersens is also about supporting divers with additional sensor technology and AI, and compensating for poor visibility. Overall, the researcher called on all rescuers to have more courage to try things out in the field of AI, despite the wide range of regulatory requirements.

(nie)