A flexible navigation system for autonomous robots operating in heterogeneous environments

Dissertação para obtenção do Grau de Mestre em Engenharia Electrotécnica e de Computadores

This dissertation presents a flexible navigation system for autonomous robot operating in heterogeneous environments. In the proposed system, flexibility occurs at several levels of the navigation system. At the lowest level, proper locomotion modes are selected according to the local context, which includes handling dynamic footprints when computing traversability costs. This flexibility ensures that the kinematic and morphological constraints of the robot are adequately considered when navigating in demanding environments. At the highest level, proper motion planning strategies are selected according to the global context, namely, the expected topology of the environment. This flexibility allows the robot to trade-off, in a context-aware way, accuracy of the planned motions and computational cost. As a result, the complexity of the planner matches the complexity of the environment, which is key to enable a proper management of computational and energetic resources. Following a pragmatic strategy, the robot obtains its global context from off-line generated maps, which are becoming widely available. To validate this idea, a tool for semantic labelling of satellite imagery was developed. Online, the robot obtains its global context extracting the semantic label (e.g.,urban environment) and distribution of expected obstacles associated to its current global position. The proposed system leverages on the well-known Robotics Operating System (ROS) framework for the implementation of the navigation system underpinnings. For extensive validation purposes, a physics-based 3D simulator was used. Moreover, the system was validated over 1 Km long experiments on a physical four-wheeled robot. The results obtained show the ability of the system to ensure safe navigation in heterogeneous environments. This is a result of the ability of the system to exploit context awareness to reconfigure itself when facing transitions among topologically different environment regions. The added level of robustness introduced with the proposed system is expected to foster the application of autonomous robots in socially relevant domains.