2021-PhD-Navigation

Social Robot Navigation using Learning Techniques

For robotic vehicles to navigate safely and efficiently in pedestrian-rich environments, it is important to correctly understand and model subtle human behaviors and common navigation rules. In the past, Model-based prediction algorithms have delivered satisfactory results [1,2]. However, once the amount of interactions and complexity of the environment increase, one should seek alternative solutions that correctly handle these challenges, such as richer expressivity of paths generated, evaluation of risk [3] or simply explainabilty of the decision making. Recently, tremendous advances have been shown in learning techniques in multiple fields. In the robotics discipline, specially on robot navigation, there have been several research works pioneering on the idea of using Gaussian Process [4], Deep Reinforcement Learning [5] or Generative Adversarial Networks [6], to name a few. Unfortunately, the improvement over classical techniques is not so spectacular as in other disciplines after using deep learning architectures.
The objective of the present project is to further push on the results obtained on human motion prediction and robot navigation algorithms.

[1] G. Ferrer, A. Garrell and A. Sanfeliu. Robot Companion: A Social-Force Based Approach with Human Awareness-Navigation in Crowded Environments. IROS, 2013.
[2] G. Ferrer and A. Sanfeliu. Anticipative Kinodynamic Planning: Multi-objective Robot Navigation in Urban and Dynamic Environments. Autonomous Robots 2019
[3] D. Mehta, G. Ferrer and E. Olson. Backprop-MPDM: Faster risk-aware policy evaluation through efficient gradient optimization. ICRA 2018.
[4] Kim, E., Choi, S., & Oh, S. “Structured low-rank matrix approximation in Gaussian process regression for autonomous robot navigation.” ICRA, 2015
[5] Chen, Yu Fan, et al. “Decentralized non-communicating multiagent collision avoidance with deep reinforcement learning.” ICRA, 2017
[6] Tai, Lei, et al. “Socially-compliant Navigation through Raw Depth Inputs with Generative Adversarial Imitation Learning.” arXiv, 2017.

Eligibility

  • Candidates should hold or expect an MSc (or an equivalent) degree in the following or closely related disciplines:
    – Robotics
    – Data Science/Computer Science
    – Engineering
  • Basic knowledge in Programming
  • Programming skills: Cpp and Python. ROS is a plus
  • High interest in Robotics.
  • Previous experience in Robotics, AI
  • Motivation to complete a PhD within 4 years