Research Fellow
University of Michigan, CSE department

2260 Hayward St, BBB office 3912
Ann Arbor, MI 48109, US
Email: gferrerm [at] umich.edu
Google scholar

Currently, I am an Intermittent Lecturer at the University of Michigan, within the Autonomy, Perception, Robotics, Interfaces, and Learning APRIL group, directed by professor Edwin Olson.

I completed my PhD thesis on robot navigation algorithms in urban environments at the Institut de Robotica i Informatica Industrial IRI, Barcelona, Spain, under the supervision of Prof. Alberto Sanfeliu. You can find here the video of the PhD defense. This work resulted finalist on the Georges Giralt Award the European award in robotics. After my PhD graduation, I moved to the University of Michigan, where I have worked as a research fellow at the APRIL Lab. and later I became intermittent lecturer, teaching the Mobile Robotics course.

Research interests

My research is focused on robot navigation, particularly my interests are within Path Planning, Perception, and Human Robot Interaction, although there are multiple topics overlapping on this field. The full autonomy milestone for robots has not arrived yet, we are slowly closing the gap, however there are still many hindrances, such as uncertainty, scalability and reliability.

During my thesis, I studied dynamic environments, highly uncertain due to the pedestrians involved. Under these conditions, complex situations arise under the interaction with people, where known path planning techniques provide poor solutions. My goal was the study and development of new prediction approaches, aiming to obtain a more intelligent robot motion behavior and introducing predictive reasoning into the planning scheme.

Later as a post-doc, I have tried to incorporate uncertainty into planning solutions, where it naturally leads to risk-aware plans. Due to the unpredictable nature of pedestrians, a perfect accuracy on prediction is not feasible. Hence, it makes sense to calculate plans on adversarial scenarios, leveraged by probability distributions, as an effective way to avoid potentially dangerous situations. Ensuring for instance an average of 99% of success rate is a great result in academia but it is a poor result if we think on long-term and safe deployment of robots. This represents the main reason why we should be interested in alternative and divergent ways to express and operate under uncertainty.