Physical AI Systems — AI with a body

Originally published by Swiss Cognitive.

In my research at KTH, I am interested in embodied AI systems – systems with a location in the world that are able to move around and also change the state of the world by means of their embodiment. An example of this is autonomous vehicles. 

An embodied AI system can actively perceive the world around them in order to gather information through sensors such as vision, process this information on different cognitive levels, both autonomously and in interaction with a human, and finally make plans or decisions based on the outcome. In the autonomous vehicle case, the internal processing consists of planning future driving actions in interaction with the surrounding world and possibly also with a driver.

Modern learning-based AI systems are extremely data-hungry; this is an issue that has to be addressed in the near future, steering development towards more data-efficient methods. An additional challenge with autonomous vehicles is that they constantly meet new situations and surroundings, which means that visual data acquisition for training the vehicle’s perception for all possible eventualities becomes infeasible. A solution is to simulate a wide variety of surroundings and generate synthetic photorealistic data; for an example of such a method from my KTH lab, see https://arxiv.org/pdf/2012.05846.pdf. 

In my role as Lead AI Scientist at Silo AI I help bring these ideas to use in industrial development projects. We are the leading private AI laboratory in the Nordics, and have broad and deep competence in real-world AI applications with a strong focus on computer vision, natural language processing and machine learning. We arrange regular workshops about various AI topics: relevant to this article is an upcoming webinar regarding autonomous systems; sign up to be the first to know more as we confirm the details: https://learn.silo.ai/webinar-autonomous-vehicles.

About the author

I am a Professor in the Division of Robotics, Perception and Learning, KTH. I am also a Lead AI Scientist at Silo AI and an affiliated researcher in the Perceiving Systems Department, Max Planck Institute for Intelligent Systems in Tübingen, Germany. I do research in Computer Vision and Machine Learning. The general theme of my research is methods for enabling artificial agents to interpret the behavior of humans and other animals, and also to behave in ways interpretable to humans. These ideas are applied in Performing Arts, Healthcare, Veterinary Science, and Smart Society. In my free time I like to play the classical double bass, and of course also spend time with the family outdoors, skiing or hiking.