Pablo Quintía

Omnivision-based KLD-Monte Carlo Localization

Mobile robots operating in real and populated environments usually execute tasks that require accurate knowledge on their position. Monte Carlo Localization (MCL) algorithms have been successfully applied for laser range finders. However, vision-based approaches present several problems with occlusions, real-time operation, and environment modifications. In this article, an omnivision-based MCL algorithm that solves these drawbacks is presented. The algorithm works with a variable number of particles through the use of the Kullback-Leibler divergence (KLD). The measurement model is based on an omnidirectional camera with a fish-eye lens. This model uses a feature-based map of the environment and the feature extraction process makes it robust to occlusions and changes in the environment. Moreover, the algorithm is scalable and works in real-time. Results on tracking, global localization and kidnapped robot problem show the excellent performance of the localization system in a real environment. In addition, experiments under severe and continuous occlusions reflect the ability of the algorithm to localize the robot in crowded environments.

Simultaneous learning of perception and action in mobile robots

One of the main problems of robots is the lack of adaptability and the need for adjustment every time the robot changes its working place. To solve this, we propose a learning approach for mobile robots using a reinforcement-based strategy and a dynamic sensor-state mapping. This strategy, practically parameterless, minimises the adjustments needed when the robot operates in a different environment or performs a different task. Our system will simultaneously learn the state space and the action to execute on each state. The learning algorithm will attempt to maximise the time before a robot failure in order to obtain a control policy suited to the desired behaviour, thus providing a more interpretable learning process. The state representation will be created dynamically, starting with an empty state space and adding new states as the robot finds new situations that has not seen before. A dynamic creation of the state representation will avoid the classic, error-prone and cyclic process of designing and testing an ad hoc representation. We performed an exhaustive study of our approach, comparing it with other classic strategies. Unexpectedly, learning both perception and action does not increase the learning time.

Combination of a low cost GPS with visual localization based on a previous map for outdoor navigation

In this work we propose the combination of a low cost GPS with a vision based localization system. Both types of localization complement each other, providing better precision and robustness than using them separately. The visual localization compares traversable regions detected by a camera with regions previously labeled in a map. We create this map by hand from images taken from Google Maps by labelling those regions passable by the robot (e.g. pavement). For the integration of both localization systems we propose the use of a particle filter. We obtained promising preliminary results taken in the surroundings of the Faculty of Computer Science of the University of A Coruña.

Implementation of robot routing approaches for convoy merging manoeuvres

Autonomous and cooperative guidance strategies for a convoy of electric vehicles in an urban context are challenging research topics in robotics and intelligent transportation systems. The vehicles that form the convoy eventually will have to leave it to perform a mission and return to the convoy formation once the mission has been accomplished. Nevertheless, the merging manoeuvres amongst the convoy and the units returning to it (pursuing units) is a complex task that involves the determination of the best merging point and the route across the city to reach it. This paper tackles this routing problem of a robot located in a map that is trying to join a convoy of robots in constant movement along a peripheral trajectory. We have developed two search strategies able to determine the optimal merging point and the best route to reach it: on one hand we describe a basic solution able to solve the problem when the time spent by the robot traveling along every street of the map is considered to be known and constant. On the other hand, we extended this basic approach to provide a new search strategy that considers uncertainty in traveling times. This increases considerably the complexity of the problem and makes necessary the inclusion of a risk factor that must be considered when determining the best route and the merging point for the manoeuvre. We also put our search strategies into practice in both, simulated and real scenarios. On one hand we have simulated the behavior of a convoy leader and a transport unit which is trying to join in the convoy, using Player&Stage. On the other hand we have used real P3-DX robot units as prototypes of electrical vehicles in a transport scenario.

Robot routing approaches for convoy merging maneuvers

Autonomous and cooperative guidance strategies for a convoy of electric vehicles in an urban context are a challenging research topic in robotics and intelligent transportation systems. The vehicles that form the convoy eventually will have to leave it to perform a mission and return to the convoy formation once the mission has been accomplished. Nevertheless, the merging maneuvers amongst the convoy and the units returning to it (pursuing unit) is a complex task that involves the determination of the best merging point and the route across the city to reach it. This paper tackles with this routing problem of a robot located in a map and that is trying to join a convoy in constant movement along a peripheral trajectory. We have developed two search strategies able to determine the optimal merging point and the best route to reach it: on one hand we describe a basic solution able to solve the problem when the time spent by the robot travelling along every street of the map is considered to be known and constant. On the other hand, we extended this basic approach to provide a new search strategy that considers uncertainty in travelling times. This increases considerably the complexity of the problem and makes necessary the inclusion of a risk factor that must be considered when determining the best route and the merging point for the maneuver. We also put our search strategies into practice, simulating the behaviour of both, the convoy and the robot trying to joining it, using Player&Stage. In this article we show the results we achieved and that validate both of the aforementioned solutions.

Learning Wall Following Behaviour in Robotics through Reinforcement and Image-based States

In this work, a visual and reactive wall following behaviour is learned by reinforcement. With artificial vision the environment is perceived in 3D, and it is possible to avoid obstacles that are invisible to other sensors that are more common in mobile robotics. Reinforcement learning reduces the need for intervention in behaviour design, and simplifies its adjustment to the environment, the robot and the task. In order to facilitate its generalization to other behaviours and to reduce the role of the designer, we propose a regular image-based codification of states. Even though this is much more difficult, our implementation converges and is robust. Results are presented with a Pioneer 2 AT. Learning phase has been realized on the Gazebo 3D simulator and the test phase has been proved in simulated and real environments to demonstrate the correct design and robustness of our algorithms.

Learning on Real Robots from Their Direct Interaction with the Environment

We present a new solution to achieve fast and continuous learning and adaptation processes on a real robot, even when the robot receives reinforcement from a human observer. The person does not need to have any kind of robotics knowledge, and will be able to provide the reward signal to the robot with a wireless joystick. Despite this highly-non-deterministic reinforcement, the robot is able to reach the desired behaviour in short periods of time.