DyNNamo, AI for the benefit of people with disabilities

With funding from the French National Research Agency, the ANR DyNNamo project aims to design AI capable of dynamically adapting to these situations. This ambitious goal can only be achieved by exploring the best compromises between software and hardware using a development method that does not prejudge solutions, but matures them by taking into account all the system's specifications.

With its sometimes confusing traffic lanes, its more or less disciplined cyclists, and its impatient motorists, crossing a major city street is already a challenge in itself for a pedestrian in full possession of their faculties.

“On the industrial side, the spontaneous tendency would be to multiply the number of AIs dedicated to each problem: one for vision, one for motor management, another for navigation or traversability, etc. But that would require far too much energy and computing power for a platform as constrained as a wheelchair,” continues the researcher. " The solution is to create an AI that is economical in terms of resources and capable of adapting to each context. And my job is to design the architecture of this embedded system. “

To do this, Jean-Christophe Le Lann is working within the framework of a paradigm called ”Dataflow," in which the embedded system is modeled as a huge flow of information at the computer chip level.

“The goal is to model the system at a certain level of abstraction and gradually descend to the hardware platform through progressive architectural refinements.”

During this process, several choices are possible, in terms of hardware or software, and it is necessary to explore them virtually to assess all the implications.

This metaphor is perfectly applicable to the field of assistance for people with disabilities, where the goal is to smooth out difficulties in order to guarantee the safest possible mobility for all.