Title:	Safe and Socially-Compliant Autonomous Robots in Dynamic Environments
DNr:	Berzelius-2026-36
Project Type:	LiU Berzelius
Principal Investigator:	Iolanda Leite <iolanda@kth.se>
Affiliation:	Kungliga Tekniska högskolan
Duration:	2026-02-12 – 2026-09-01
Classification:	10210
Keywords:

Abstract

This research project brings together three PhD students (two of them WASP-funded and another one working on a WASP-funded project) and their research engineers working on Human-in-the-Loop Reinforcement Learning (RL) and human-aware dynamic scene representations. Whether it is adjusting a driving policy for the unique characteristics of a truck, ensuring that a robot’s reward model is transparent to human supervisors, or enabling a robot to reason about human routines in a 3D space, we focus on data efficiency and interpretability. By leveraging techniques such as Low-Rank Adaptation (LoRA), physics-informed RL, and leveraging LLM to improve robots’ perception of social environments, we aim to create autonomous agents that are not only capable but also compliant with humans and physical constraints.