TLDR : Zurich-based start-up Lightly presents LightlyEdge, an embedded AI solution that selects and transmits only relevant data from vehicle cameras and sensors. This method could offer European car manufacturers a way to reduce costs and accelerate development, providing a competitive edge against AI leaders.
Table of contents
As European car manufacturers face intense pressures (technological competitiveness, workforce reductions, and geopolitical turbulence), Zurich-based start-up Lightly announces the launch of LightlyEdge, an AI-based data collection solution. The goal: to sort data right from capture, transmitting only the essentials without sacrificing the quality of AI model training.
Embedded AI: Capture Less, Capture Better
LightlyEdge aligns with the growing trend of edge AI, which involves running models directly on devices, in this case, the cameras and sensors of vehicles. Instead of continuously recording every kilometer, the system analyzes video streams in real-time and selects only the scenes deemed relevant: hazardous situations, rare weather conditions, unexpected behaviors.
This source filtering addresses a well-identified constraint in the industry: the explosion of data volume has not always equated to improved model performance. On the contrary, it results in increasingly high storage, transfer, and processing costs while sometimes diluting the rare and critical cases that could enhance the diversity and quality of training datasets.
A Strategic Asset for European Manufacturers?
While Tesla has mastered active learning loops based on selective collection for years, European manufacturers struggle to keep up, hindered by more rigid architectures and increased reliance on third parties for data management.
By transposing this logic to an edge solution compatible with existing fleets, LightlyEdge addresses two challenges: cost reduction and acceleration of the development cycle. According to Matthias Heller, co-founder of the start-up, "With LightlyEdge, our partners can leverage smarter, real-time data collection that not only speeds up AI model training but also provides a competitive advantage over established industry giants."
For European players, still marked by an engineering culture focused on system performance, this refocusing on data relevance represents a breakthrough. But it is perhaps in this breakthrough that lies the possibility of regaining ground against native AI champions.
Translated from LightlyEdge : une IA embarquée pour alléger la charge des données dans la course à la voiture autonome
To better understand
What is embedded artificial intelligence and how does it work in the context of vehicles?
Embedded artificial intelligence (edge AI) runs AI models directly on devices like vehicle sensors. It processes data in real-time on the device, allowing it to filter out relevant information before sending it to central servers, reducing bandwidth and storage requirements.
What are the regulatory challenges related to the use of embedded AI in vehicles in Europe?
Embedded AI in vehicles raises regulatory issues concerning data privacy, safety, and liability in case of accidents. In Europe, the GDPR imposes strict data collection requirements, and vehicle safety regulations need to evolve to encompass these new technologies.