Overview

MSCA-DN TRAIL (TRAnsparent, InterpretabLe Robots) is a Horizon Europe Marie Skłodowska-Curie Action Doctoral Network focused on a novel, highly interdisciplinary research and training program. The project aims to improve transparency in deep learning, artificial intelligence, and robotics systems.

TRAIL trains a new generation of Early Stage Researchers (ESR) to become experts in designing and implementing transparent, interpretable neural systems and robots that can interact effectively and safely with humans.

Objectives

1. Behaviour Transparency: Enabling robots to communicate their internal states, intentions, and planned actions through intuitive channels like movements, gestures, and audio-visual cues.
2. Decision Transparency: Developing methods to interpret deep neural networks and extract knowledge from their decision-making processes.
3. Interdisciplinary Training: Combining expertise in AI, robotics, computer science, human-robot interaction, and psychology to create a comprehensive training curriculum.
4. Human-Centric Design: Ensuring that transparency mechanisms are tailored to human cognitive abilities and social expectations.

Research Areas

Behaviour Transparency

• Multimodal robot communication
• Socially expressive robotic behaviours
• Legibility of robot intent

Decision Transparency

• Explainable AI (XAI) for robotics
• Interpretable deep neural architectures
• Knowledge extraction from neural models

Human-Robot Interaction

• User studies on robot transparency
• Social cognition in robotics
• Evaluation of interpretable systems in the wild

Technology Stack

• Deep Learning frameworks (PyTorch, TensorFlow)
• Explainable AI techniques
• ROS 2 for robot control and communication
• Human-Robot Interaction evaluation tools
• Advanced perception and multimodal sensing

Expected Impact

• Increased Trust: Enhancing human trust in AI and robotic systems through better understanding.
• Safety and Reliability: Improving the safety of autonomous systems by making their decisions more predictable.
• Technological Advancement: Pushing the boundaries of interpretable AI and socially aware robotics.
• Expertise Development: Training 15 ESRs as the future leaders in transparent and ethical AI.

Current Status

TRAIL kicked off in early 2023 and is currently in its active research and training phase. The Doctoral Network is actively conducting research across its partner institutions and providing specialized training events for its researchers.

Key Publications

Dynamics of Mental Models: Objective Vs. Subjective User Understanding of a Robot in the Wild

Anaís Garell , Ferran Gebelli , Ferran Gebellí , Raquel Ros , Séverin Lemaignan

2025 - IEEE Robotics and Automation Letters DOI: 10.1109/LRA.2025.3579217 Social Robotics