Doctoral researcher in Situationally-Aware Robot Motion Planning and Control using Probabilistic and Learning-based Approaches

The SnT Automation & Robotics Research Group seeks to hire an excellent and motivated PhD candidate within the national research project PCS-GRAPHS (Integrating Situational Awareness, Planning and Control for Autonomous Robots using S-Graphs), funded by the Luxembourg National Research Fund (FNR).

The successful candidate will work under the main supervision of Prof. Holger Voos and will be required to perform the following tasks:

• Carry out the aforementioned research with national and international collaborators, e.g., INRIA Rennes, University of Zaragoza, TU Munich or University College London
• Disseminate your findings at renowned international robotic conferences and workshops such as e.g., ICRA, IROS, and in top-level journal papers
• Supervision of Master and Bachelor students contributing to the project
• Provide assistance in organizational matters related to the project PCS-Graphs

Since autonomous robots need to operate in complex and ever-changing environments for long periods, they must constantly understand their surroundings to make smart decisions and complete tasks, such as moving safely without collisions or handling objects. Recent methods for robotic situational awareness (SA), like our work on Situational Graphs (S-Graphs), improve on existing techniques by combining 3D environmental maps with detailed knowledge about objects into a single, optimized model. First novel solutions use parts of these models for planning or control, but they do not take full advantage of the structured, layered information such graphs so far. Therefore, our project aims to tightly integrate S-Graphs with motion planning and control.

For that purpose, we will first extend our SA models to XS-Graphs by including additional information necessary for motion planning and control, representing the situation with Probabilistic Graphical Models such as Factor Graphs based on the fusion of multimodal sensorial information and integration at different layers of abstraction. The major contribution of the PhD will be the derivation of multilayered approaches for motion planning and control based on the XS-Graphs, where both model-based and learning-based solutions are foreseen. This includes the development of a higher-layer semantic planning, the decomposition of the plan in subsegments and dedicated controller (such as MPC) for combined planning and control on a subsegment of the XS-Graph. This exploitation of the multilayered structure of the XS-Graphs will lead to a very efficient planning and control approach to outperform SOTA solutions. The overall approach will be tested and assessed in simulations, experiments and use case demos using our laboratories equipped with drones, legged and humanoid robots.

For further information, please contact Holger Voos at: holger.voos@uni.lu