Robotic Exploration

For three decades, through research and investigation with robots including Ambler, Dante, Nomad, Kambara, Hyperion, Zoë, Scarab, and others, I have evaluated and proven new concepts, technologies and robots in extreme environments around the globe.  I see the movement of robots into the field and the embodiment of greater autonomy in machines as the natural evolution of robotic science. 

The practical and economic reasons for using robots to explore the planets, moons and asteroids are clear: the risks are high, resources are scarce, and communication is limited.  To answer profound questions about the past, present and future of life and the nature of our solar system, effective robotic systems are crucial.  

In underwater environments, technologies for visual servoing, sonar-based localization and mapping, and learned control can transform the safety and efficiency of operation.  In air, atmospheric studies will develop from periodic sampling to perpetual robotic presence for monitoring and investigation.  On Earth, in remote and hazardous regions where both precision and endurance are required, there are myriad applications for robotic explorers. 

I expect that science robots will become commonplace, operating on the land, sea and air of the Earth and in space engaging in scientific discovery and meaningful work.  My aim is to discover the knowledge and techniques needed to realize the full potential of robotic exploration. I am concentrating on several areas of research that are fundamental: endurance navigation, robust autonomy, and scientific data understanding.

Endurance Navigation

Robots that explore must endure over long distances and durations; they must travel over the horizon, and be able to map, localize, and navigate in terrain that is unknown in detail and only sparsely observed.

In my work enabling robots to map the distribution of microorganisms in desert environments, the ability to perform multi-kilometer autonomous traverse is essential.  This new capability for robotic survey pioneers a method of science investigation that reveals regional as well as local characteristics.  I formulated the robotic survey approach and have fostered the development of endurance navigation that has now proven the feasibility of single-command, multi-kilometer traverse and resulted in over 250 kilometers of autonomous survey in the Atacama Desert of northern Chile.

How do we enable robots to sustain efficient exploration?  Exploration by its nature involves uncertainty.  This must be addressed in a variety of aspects.  My continuing navigation research dynamically accounts for uncertainty in localization at multiple scales by maintaining topological structures of metric observations: satellite overviews, far-field perception, and near-field terrain models. Dynamic features and time-varying aspects of the environment are also detected and modeled.  As localization is refined, multi-scale information is registered and fused to improve terrain evaluation.  Perception over distances, in the far-field and from remote sources, is needed for endurance. I am working on methods of applying sparse information in the distance to better inform immediate navigation decisions.  Eliminating any gap between what robots can perceive locally and what they know globally is vital for robots to explore efficiently.  

Robust Autonomy

Reasons for autonomy derive from communication bandwidth and latency: robots generate far more information than they can transmit and can get into trouble much more quickly than they can expect to receive assistance.  Another strong motivation for autonomy is the cost and risk associated with keeping humans too long at the controls.  I believe it is more reliable and economical to allow robots to act autonomously. 

Resource-cognizant planning enables robots to reason about available resources. For example, a solar-powered robot must choose actions that provide sufficient energy by avoiding shadows and orienting in synchrony with the sun.  Field experiments that I led in the Canadian arctic proved the concept of sun-synchronous exploration by enabling robots to manage their energy balance while reaching science goals.

My work in underwater exploration involves enabling a robot to autonomously map and sample flooded cave systems in central Mexico.  The investigation demanded resource-cognizant planning with even greater consideration of consumable resources, like battery energy, pressurized gas, and sample storage, and with more diverse activities including instrument observations and sampling operations.   Applying specializations of simultaneous localization and mapping the robot was able to explore and reach the bottom, at 318 m, of the world’s deepest cenote.  The innovations include algorithms to recognize location by geometric structure rather than specific features, for contingent plan execution for fault recovery, and to efficiently model fully three-dimensional environments for evaluating multiple hypotheses of map and location.  I am applying these techniques to other feature-poor natural environments.  

What level of autonomy is needed for robotic exploration? My ongoing research is extending the robotic exploration architecture by furthering fault detection and diagnosis into recovery and incorporating contingent planning. Robustness improves as robots are more able to analyze their own performance, detect faults, correct actions, and learn to improve. I will continue to evolve architectures that support robot self-awareness through detailed state monitoring and reasoning about resources, faults, and recovery actions. 

How should we conduct remote investigations? New techniques for autonomous exploration paradoxically introduce the need to examine human-robot interaction and to establish a methodology of remote investigation.  In order to define effective methods of remote investigation, I am collaborating with planetary scientists to study mission operations and establish an understanding of how scientists relate to robot explorers, how they can guide robots most effectively, and how this affects the quality of scientific results.

Understanding Science

Robotic explorers, while not able to formulate hypotheses or draw conclusions from their investigations, must act autonomously to seek, measure, record, verify and reduce information.  Science autonomy means enabling robots to reason about scientific objectives and methods, or more explicitly, to possess a model of science objectives, a method of evaluating observations, and the ability to modify exploration behavior and plans.

Science-on-the-fly is a concept that is required when robots are able to travel over the horizon and beyond frequent human oversight: the robot must decide what measurements and observations to make and where exactly to make them. During traverse between investigation sites, a robot with no ability to interpret its observations on the fly, would drive blindly past a dinosaur fossil.  An explorer must be alert to novel observations as well as differences in  feature classification and distribution that indicate changing conditions.  In field experiments in Amboy, California, the learned classification of spectrometer observations and evaluation of model entropy was used to dynamically adjust path to maximize information gain.  This resulted in the first automatic generation of geologic site maps.

I am investigating techniques for detection and classification in the domain of natural features.  Identifying rocks, soils, and geologic boundaries requires diverse methods of image analysis and probabilistic classification.  Working with imagery from the Atacama desert, or from Mars, robots may now automatically count particle (rock) size and distribution, distinguish regional characteristics, and make decisions about geologic boundaries that can be used to inform exploration strategy and distribute observations in ways that maximize efficiency and value.

An important outcome of this research has been the development of algorithms for visual detection of natural features, like rocks, and evaluation of their informational content.  This work has informed the development of software now operating on the Mars Rovers to find significant rocks in panoramic views and include high-resolution images in the daily report cycle.

How can robots understand and act on the goals of a scientific investigation?  Work in science autonomy will improve the effectiveness by which instrument data can be analyzed, reduced and conveyed to scientists.  I envision robotic explorers with sufficient modeling of geologic, biologic, and natural processes that they can intelligently map regions, making the most effective observations, while using fewer resources.  This will dramatically improve the quality of their scientific results.  The embodiment of science autonomy will transform robots from automated vehicles into true explorers.

Field Experimentation

The deployment of robots in challenging environments is the unifying and sustained theme in my work, it motivates the systemic view that I take on research and it leads to field experimentation, which is essential to advancing robots as tools for science. 

Creating field robots is an endeavor in which systems engineering is a fundamental concern.  Robots are the very definition of a complex, engineered system.  The complexity is such that engineering principles and practice that guide design, analysis, and synthesis are slow to be established.  Most robotics research never escapes the laboratory to be validated in the field.  Through thoughtful and repeated development and evaluation, effective methods of engineering robotic systems are emerging.  I continue to confront the practical challenges of creating field robots and work to clarify and teach best practice in systems engineering. To do this I collaborate across disciplines with engineers and scientists, and train students and teach courses in the design and development of robotic systems.

I am convinced that field experimentation provides necessary verification and validation of research progress, as well as essential grounding and motivation.  It is by fielding systems and making discoveries that we can prove the potential and importance of robotic exploration.