Research

In Computational Robotics laboratory (ComRob) we perform fundamental and applied research at the intersection of the artificial intelligence and autonomous robotic systems. We are seeking for unique solutions to address real world challenges to improve quality of life and to understand principles emerging in nature.

We work on demanding problems of planning and control of complex robotic systems in autonomous data collection that require a deployment of new adaptive and learning techniques. Our focus is not limited to build remotely-controlled robots, but we further develop advanced techniques for efficient solution of problems related to environmental monitoring, surveillance, and reconnaissance missions.  

Current projects

  • Robotic Lifelong Learning of Multi-legged Robot Locomotion Control in Autonomous Data Collection Missions, Czech Science Foundation (GA ČR), project No. GA18-18858S (2018-2020)
  • Efficient Information Gathering with Dubins Vehicles in Persistent Monitoring and Surveillance Missions, Czech Science Foundation (GA ČR), project No. GA16-24206S (2016-2018)
  • Hybrid navigation system for autonomous vehicles in environment with denied GNSS services, Technology Agency of the Czech Republic (TA ČR), project No. TH03010362 (2018-2020)

Past projects

  • Adaptive Informative Path Planning in Autonomous Data Collection in Dynamic Unstructured Environments, Czech Science Foundation (GA ČR), project No. 15-09600Y (2015-2017)
  • Self-Organizing Maps for Multi-Goal Path Planning Tasks, Czech Science Foundation (GA ČR), project No. 13-18316P (2013-2015)

Current projects

Robotic Lifelong Learning of Multi-legged Robot Locomotion Control in Autonomous Data Collection Missions

The proposed project aims to contribute to the field of autonomous data collection planning by developing new robotic lifelong learning techniques and on-line planning algorithms with theoretically and experimentally established properties. Within the project, we will investigate self-supervised incremental on-line learning algorithms for terrain learning and traversability assessment together with locomotion control to generalize existing frameworks of autonomous data collection planning towards online adaptive planning based on the robot’s experience with traversing the terrain. Hexapod walking robots are planned to be used in the studied data collection missions which allow traversing uneven terrains at the cost of complex and computationally demanding precise motion planning. Therefore, the project focuses on developing novel efficient methods and models of locomotion control policies to enable online adaptive robot control in challenging terrains and thus improve the performance of autonomous data collection during the mission.

 

Efficient Information Gathering with Dubins Vehicles in Persistent Monitoring and Surveillance Missions

The proposed project aims to contribute to the field of robotic information gathering with non-holonomic (Dubins) vehicles by developing planning algorithms with theoretically established properties and their experimental validations. We plan to leverage on analysis of the paths with bounded curvature modeled as optimal maneuvers for Dubins vehicle to establish new approximation bounds for Dubins planning. We propose to develop novel discretization techniques to reduce the Dubins traveling salesman problem with neighborhoods to discrete combinatorial optimization problems that can be address by existing efficient combinatorial heuristics to enable a direct solution of the orienteering problem with Dubins vehicles. In particular we aim to : 1) establish approximation bounds for efficient solution of the Dubins traveling salesman problem with neighborhoods; 2) develop novel planning algorithms for persistent monitoring and surveillance missions; 3) establish complex analysis and empirical evaluation of the develop solutions in realistic experimental scenarios.

Past projects

Adaptive Informative Path Planning in Autonomous Data Collection in Dynamic Unstructured Environments

In this project, we aim to develop new adaptive planning algorithms for robotic information gathering in unstructured environments. We plan to leverage on combination of active sensing, planning, and learning techniques to design a unifying adaptive autonomous data collection planning framework to deal with motion and sensor uncertainties in a dynamic and partially known environment. We propose to generalize the autonomous data collection planning to improve performance of the data collection mission by adaptive re-planning based on new information gathered during the mission.

In particular we aim to propose and design: 1) specific autonomous data collection planning algorithms for simultaneous determination of sensing locations and trajectory generation; 2) adaptive planning algorithm for on-line refinement of traversability cost in dynamic rough terrains for a hexapod walking robot; and 3) to establish complex analysis and empirical evaluation of the developed solutions.

Self-Organizing Maps for Multi-Goal Path Planning Tasks

The project aims to investigate principles of self-organizing maps (SOM) for routing problems in high-dimensional configuration spaces. The addressed problems are from the family of multi-goal path and motion planning problems, where an optimal path or trajectory connecting a given set of goals has to be found. The motivation of the proposed scientific effort is to find a path/trajectory regarding realistic capabilities of mobile robots including their autonomous navigation. Thus, a found path would improve reliability and robustness of the autonomous navigation enabling robotic applications in daily tasks. The project comprises three scientific objectives. The first one deals with finding a representation of high-dimensional configuration space allowing usage of SOM principles for planning in an effective way. The second objective aims to develop a multi-goal motion planning algorithm providing a local planner and estimation of distance metric in the space. Finally, the third objective is to establish the planning framework considering localization uncertainties.