An extension of an existing benchmark for evaluating Vision Language Action Models in simulation that is more comprehensive, faster, and easier to extend.

An LLM-based natural language control interface for Pollen Robotics' Reachy robot, allowing users to control the robot through voice. A user study compares it against VR teleoperation.

A model that generates music aligned to a given video by fusing optical-flow motion cues with CLIP-based visual embeddings to condition an autoregressive music decoder.

A holistic view of reactive shield synthesis that unifies existing literature and provides a complete formal proof for the construction of winning regions in safety g.

Explores LLM-based in-context learning to detect imposter emails in clinical study recruitment, evaluating prompt design, few-shot setups, and weak supervision strategies.

A demonstration that current LLMs (late 2024) are still susceptible to persuasion-based jailbreaking, along with an investigation into defense strategies.

Proposes two scheduling policies for Cyber-Physical Systems in dynamic environments that maximize processor utilization while maintaining a safety margin.

An investigation into how LLM Agents can acquire new skills from a set of primitive actions that help to solve tasks in a specific environment.

A proof of concept for decentralized identities in the context of car rental, consisting of microservice-based applications that issue and verify digital driver licenses.

An in-depth analysis of decentralized identity protocols (OpenID4CI, OpenID4VP, SIOPv2, OAuth 2.0, OIDC) and a discussion of their applicability in a proof of concept.