Use this URL to cite or link to this record in EThOS:
Title: Lifelong information-driven exploration for mobile robots to complete and refine spatio-temporal maps in changing environments
Author: Machado dos Santos, Joao Pedro
ISNI:       0000 0004 6495 3660
Awarding Body: University of Lincoln
Current Institution: University of Lincoln
Date of Award: 2017
Availability of Full Text:
Access from EThOS:
Access from Institution:
Recent improvements in the ability of mobile robots to operate safely in human populated environments have allowed their deployment in households, offices and public buildings, such as museums and hospitals. However, the structure of these environments is typically not known a priori, which requires the robots to build their own models of their operational environments. This process is commonly known as "exploration" in mobile robotics. Moreover, real-world environments tend to change over time, which means that to achieve long-term autonomous operation, robots must also update their environment models as a part of their daily routine. The assumption of a perpetually-changing world adds a temporal dimension to the exploration problem, making exploration a never-ending lifelong learning process. To the best of our knowledge, this process termed "lifelong exploration" has never been studied in detail before and forms the main topic of the work presented in this thesis. Effcient lifelong exploration requires a robot to choose the right locations and times at which to collect observations in order to improve its environment model. To evaluate the ability of a robot to build and maintain its environment models, we need to be able to compare lifelong exploration strategies under repeatable experimental conditions. An evaluation methodology based on pre-recorded sensory datasets would not be suitable for this purpose, as this would not allow the robot to choose the location or time of its observations. Evaluating lifelong exploration requires the deterministic reproduction of environment changes, while preserving the robots ability to decide upon its own actions during the experiment. This thesis therefore contributes a new benchmarking methodology for lifelong exploration, which replicates the events occurring in real environments through physical simulations that reflect the environment changes gathered by ambient sensors over long periods of time. The established experimental benchmarks are based on long-term sensory datasets recorded in a smart home, with dynamics produced by a single person over a period of one year, and an office environment, with dynamics produced by a team of workers. Building upon the aforementioned benchmarking methodology, the thesis investigates possible strategies for lifelong exploration. An experimental comparison of lifelong exploration strategies that combine various decision-making paradigms and spatio-temporal representations is presented. Moreover, a new approach to lifelong explorations is proposed that applies information-theoretic exploration techniques to environment representations that model the uncertainty of world states as probabilistic functions of time. The proposed method explicitly models the world dynamics and can predict the environment changes. The predictive ability is used to reason about the most informative locations to explore for a given time. A 16 week long experiment shows that the combination of dynamic environment representations with information-gain exploration principles allows to create and maintain up-to-date models of continuously changing environments, enabling efficient and self-improving long-term operation of mobile service robots. The final part of the thesis considers the problem of acquiring and maintaining dense 3D models of dynamic environments during long-term operation, building on the work presented in the earlier chapters. The term "4D mapping" is used to indicate 3D mapping by mobile robots over extended periods of time. A new approach to lifelong 4D mapping and exploration is presented, which was deployed on a real robotic platform during long term operation in real-world human-populated environments. The approach integrates sensory data captured by the robot at different times and locations into a global, metric I 4D spatio-temporal model and then uses the model to decide where and when to perform the next round of observations. Finally, the deployment of the 4D exploration method in a real-world office scenario is described and evaluated. The one week long experiments show that the method enables reliable 4D mapping and persistent self-localisation of autonomous mobile robots, continually improving the robots maps to reflect the ever-changing world.
Supervisor: Duckett, Tom ; Hanheide, Marc ; Krajnik, Tomas Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: H671 Robotics