EdgeAI-Trust | Trustworthy edge AI for safety-critical applications

Building trustworthy artificial intelligence where data is generated 

Edge AI is rapidly emerging as a key enabler for intelligent and autonomous systems that must operate in real time, often with limited connectivity and strict safety requirements. From autonomous vehicles to industrial robotics, the ability to process and interpret sensor data locally, without relying on cloud-based processing, is becoming a strategic necessity. EdgeAI-Trust develops a European ecosystem for trustworthy edge AI, enabling intelligent devices to collaborate, analyse data and make decisions closer to where information is generated. 
 

Target group

The project is relevant for organisations developing or deploying AI-driven systems that require real-time performance, reliability and safety, including:

• Automotive OEMs and Tier-1 suppliers
• Industrial automation and robotics companies
• Mobility and intelligent transport system providers
• Semiconductor and embedded system developers
• AI software and toolchain providers
• Research organisations and standardisation bodies

Context

The number of sensors embedded in vehicles, machines and infrastructure is growing rapidly. Cameras, radars, LiDAR systems and connected devices continuously generate large volumes of data that often need to be analysed within milliseconds.

Sending all this information to the cloud is not always practical. Latency, bandwidth limitations, privacy concerns and safety requirements increasingly demand that data processing takes place closer to the source.

At the same time, deploying AI on distributed and resource-constrained devices raises important challenges. Systems must remain reliable, secure and energy-efficient while operating across heterogeneous hardware platforms. New approaches are therefore needed to validate, orchestrate and optimise AI applications running across decentralised edge infrastructures.
 

Objectives & results

EdgeAI-Trust aims to develop a domain-independent ecosystem that enables trustworthy AI deployment across heterogeneous edge devices and distributed infrastructures.

Project objectives:

• Develop a reference architecture for decentralised edge AI systems
• Enable secure, fault-tolerant and real-time collaboration between heterogeneous edge devices
• Develop AI technologies for distributed learning, sensor fusion and on-device inference
• Create tools for optimisation, validation and energy-efficient deployment of edge AI applications
• Establish the EDEM platform to monitor ecosystem performance, trustworthiness and sustainability
• Validate the developed technologies in safety-critical use cases, including autonomous mobility
• Contribute to standardisation and drive exploitation across European industry

Key expected results

• A modular edge AI architecture supporting deployment from embedded devices to cloud infrastructures within a zero-trust framework
• AI components for distributed intelligence, real-time perception and collaborative decision-making
• Methods for validating and benchmarking decentralised AI systems
• Energy-efficient orchestration techniques for heterogeneous AI workloads
• Demonstrators in automotive, mobility and industrial environments
• Open datasets, benchmarks and validation methodologies supporting future innovation and standardisation efforts

These results will help European industry accelerate the adoption of trustworthy AI solutions while strengthening technological sovereignty in strategic sectors.
 

Approach

Within EdgeAI-Trust, Sirris develops and validates decentralised AI technologies for intelligent transportation systems, working closely with industrial partner IVEX. The focus is on transforming raw sensor data into structured and actionable knowledge that can support the development and validation of autonomous and connected mobility solutions.

1. Multi-sensor perception at the edge

Sirris combines data from multiple on-board sensors, including cameras and radars, to create a comprehensive understanding of the vehicle's environment. Real-time perception algorithms detect and track road users, identify lane markings, analyse weather conditions and recognise other contextual information relevant to driving.

2. Driving scenario understanding

Building on this perception layer, AI models analyse how road users interact over time. This makes it possible to automatically identify meaningful driving situations such as lane changes, overtaking manoeuvres, intersection crossings and other traffic events that are important for testing and validating intelligent transportation systems.

3. Searchable driving-scene knowledge

The detected events are transformed into structured datasets that support both real-time analysis and large-scale offline evaluation. Sirris further develops vision-language AI technologies that allow users to search extensive video archives using natural-language queries. This enables the efficient retrieval of rare, critical or complex driving scenarios without the need for manual annotation.

4. Validation of trustworthy edge AI

The developed methods contribute to the validation and optimisation of decentralised AI systems operating on heterogeneous edge devices. By combining perception, scenario understanding and semantic search, the project helps build more reliable, explainable and trustworthy AI solutions for safety-critical mobility applications.
 

Funding

• Funding agency: European Union
• Project type: Horizon Europe – Chips Joint Undertaking (Chips JU)
• Contract number: 101139892

Official logo of the project

Partners

EdgeAI-Trust brings together 51 partners from 13 European countries: Germany, Austria, France, Italy, Spain, Denmark, Belgium, Turkey, Greece, Latvia, Cyprus, Slovenia, Switzerland.

The list of partners includes: NVIDIA, STMicroelectronics, AVL, ZF, DLR, BSC, ZHAW, UPV, IVEX. The full list can be found on https://www.edgeai-trust.eu/consortium/