A New Age for Underwater Autonomy
By Kira Coley
It is an exciting era for ocean exploration in a time when several scientific fields are beginning a new phase of discovery through advancements in underwater technology. While the use of autonomous underwater vehicles (AUVs) have been key to propelling scientific investigation forward, the complexity, time and resources needed to plan these missions have begun the quest of ‘true autonomy’ by engineers in the ocean communities. Now a new programming approach developed by engineers at MIT offers AUVs more “cognitive” capabilities, enabling these systems to form their own mission plan with minimal input from their human counterparts.
The Role of Engineers in AUV Missions
Since the 1960s, AUVs have been invaluable in scientific investigation and discovery, giving researchers access to some of the world’s most remote, and often dangerous, underwater locations. While a proven tool in oceanic research, AUV missions are complex, costly and time-consuming, requiring an expert team of scientists and engineers to safely guide the AUV while capturing the all-important scientific data.
Although the mission objectives are defined by the scientist, it is the engineer who communicates to the AUV what order each site should be visited, how each location is surveyed and which route is safe for the vehicle. This mission schedule can be affected by a number of environmental conditions, such as the effect of tide on currents, and environmental features such as reefs, mounts and other vehicles which increase the risk of collision.
Being central to the missions’ success, engineers must consider trades between the growing uncertainty when navigating under the ocean’s surface, against the cost and benefit of surfacing to communicate with satellites - an often costly process using both valuable time and energy resources. Occasionally scientific objectives may need to be sacrificed so the mission can be completed within the given timeframe and energy margins.
“Currently, the scientists will tell an engineer their science goals, priorities and additional constraints, such as when to examine different science areas, and what to avoid. The engineer uses his or her knowledge of the vehicle, plus a map of the area to come up with a sequence of low-level commands (called a script) to tell the vehicle where to go, and what to do. Commands include ‘follow a straight line to a way point,’ ‘surface,’ and ‘establish a satellite link.’ For the new generation of small vehicles, like the Slocum glider, the same person might be playing both the role of scientist and engineer,” says Brian Williams, a professor of aeronautics and astronautics at MIT.
While these robots are effective at carrying out low-level tasks specifically assigned to them by human engineers, it is a tedious and time-consuming process. For years engineers have been working on improving the autonomy of these underwater vehicles in an effort to reduce the level of human control and provide AUVs with robust decision-making capabilities.
A New System for Underwater Automation
Professor Brian Williams is the principal developer of a new mission-planning system for AUVs, developed by MIT engineers in collaboration with Dr. Richard Camilli’s team at the Deep Submergence Lab at the Woods Hole Oceanographic Institute (WHOI). The new programming approach gives robots more “cognitive” capabilities, enabling humans to specify high-level goals, while a robot performs high-level decision-making to figure out how to achieve these goals.
Williams explains, “We have developed a system in which the AUV is given the scientist’s goals directly. The vehicle then automatically generates the script of commands that the engineer would normally construct by hand. This requires the AUV to search through a large space of possible scripts for the one that it determines is best.”
Mission planning and navigation are repeatedly performed as coordinated tasks online to form a mission plan based on updated information on the vehicle status, environmental conditions and mission progress. Each time the AUV communicates an update by satellite it revises its plan based on the new information, and begins executing a new script. This allows the vehicle to adapt to potential dangers and exploit potential opportunities that occur along the way.
Over the last decade, one of the key challenges for AUV engineers has been creating a system that can automatically search through a large set of plans for one that is appropriate for its mission objectives. Traditionally, autonomous systems either avoid the ability to search through options, are constrained to explore a small set of options, or take hours to form a suitable plan. Modern planners can now search through a very large set of options in seconds. Another challenge is safely navigating areas of scientific interest which are also dangerous to the vehicle, such as reefs. This requires planners that can reason about uncertainty and risks to ensure that they leave an appropriate safety margin, and planners that replan continuously.
This new system allows the robot to plan out a mission, choosing which locations to explore, in what order, within a given timeframe - a process usually determined by the engineers. The system also plans how to safely and efficiently navigate the vehicle between and within the science areas. This includes deciding how close to get to parts of the reef, given the uncertainty of currents, and when to surface, in order to gain a better position estimate. If an unforeseen event prevents the robot from completing a task, it can choose to drop that task, or reconfigure the hardware to recover from a failure.
In March this year, the MIT engineers, along with groups from WHOI, the Australian Center for Field Robotics and the University of Rhode Island, tested the new mission-planning system on an autonomous underwater glider during a research cruise off the western coast of Australia.
“The trials went even better than expected. We started developing the capability in layers and adapted them each day based on our experiences,” recalls Williams. “First, Rich Camilli’s team worked to ensure that the Slocum glider was functioning properly, was well calibrated and reliable. We then added reasoning capabilities in layers. Several different approaches to navigation was tested, until we found one that managed risk and energy effectively. We then incorporated the ability to monitor the environment, and to replan the vehicle routes as information was updated. This process was repeated for the mission planner. Finally, we introduced more complex science goals, where the vehicle was asked to explore areas that were normally considered too dangerous for gliders operated using traditional methods.”
The Slocum glider, using this system, was able to adapt its mission plan to avoid getting in the way of other vehicles, while still achieving its most important scientific objectives. If another vehicle was taking longer than expected to explore a particular area, the glider would reshuffle its priorities, and choose to stay in its current location longer in order to avoid potential collisions. After a week of autonomously operating the glider, the team found that the new approach successfully enabled the vehicle to perform science in areas that were previously only possible using more costly, traditional AUVs.
“I don’t think in terms of ‘true autonomy’. I view the future as a partnership between human scientists and autonomous vehicles.”
“Over time these vehicles will become increasingly able to evaluate a progressively large set of options, given what they have been told. In this way they are like a new form of calculator. But the vehicle plans are only as good as the model that the scientists and engineers give them,” says Williams. “Humans will guide the vehicles in the form of goals, priorities and safety constraints. They will need to tell the vehicles what is important, what risks are acceptable, and what uncertainties to model.”
Autonomy in Sea and Space
The autonomous mission-planning system, named Enterprise after the fictional starship in the “Star Trek” franchise, is similar to one that Williams developed for NASA following the loss of the Mars Observer, a spacecraft that, days before its scheduled insertion into Mars’ orbit in 1993, lost contact with NASA. Based at NASA’s Ames Research Centre, Williams at the time, was tasked with developing an autonomous system that would enable spacecraft to diagnose and repair problems without human assistance. The system was successfully tested on NASA’s Deep Space 1 probe, which performed an asteroid flyby in 1999.
“An autonomous underwater vehicle can be thought of as a cross between a rover or deep space probe, and an autonomous air vehicle.”
“Like space exploration, the underwater vehicles are performing science in an unreachable area, and with limited communication. We talk to a Mars Rover once a day and we talk to the autonomous underwater vehicle only when it goes to the surface. Like an air vehicle, the underwater vehicle flies through the ocean, while being buffeted by currents. The uncertainty that results from this dynamic environment is a very significant issue,” explains Williams.
“A key difference between ocean and space, is that the most complex space missions, like Cassini and Curiosity, have an enormous amount of redundancy, and a large number of science instruments. Cassini and Curiosity both cost in excess of a billion dollars, so it’s hard to take risks with these vehicles, hence Mars rovers today use very limited autonomy. Our work at NASA focussed on systems that could diagnose and repair these space systems automatically. The underwater vehicles are much simpler internally, so our focus is less on diagnosing and repairing internal hardware, and more on navigating the ocean safely.”
Future of Underwater Autonomy
Autonomous vehicles are already being used in support of a very broad range of missions including monitoring reefs and fisheries, deep sea exploration, archaeology, oil exploration, disaster response and security. By giving robots control of higher-level decision-making, engineers are free to think about overall strategy, while AUVs determine for themselves a specific mission plan. This new ability could also reduce the size of the operational team needed on research cruises and enable robots to explore places that otherwise would not be traversable.
As we enter this new age of underwater autonomy, AUVs will provide ways to persistently monitor large areas in a much more cost effective manner. Today, a cruise typically operates a single autonomous vehicle, with costs in excess of a million dollars, performing limited duration missions. In contrast, vehicles within the realm of $100K can be deployed by two people by hand, operate for weeks at a time, and require less time and expertise to command. With this relatively low cost, a future in which many vehicles are performing science exploration cooperatively can be anticipated.
“Our friends who are AUV operators and Mars rover drivers both complain that they spend too much time generating low-level command sequences, and do not have enough time to think strategically, either in terms of the science or engineering contingencies. We hope that our tools offer them more time to spend focussed on what they want to do, and are uniquely qualified to perform. Therefore this system could shape a new role for engineers so that they are able to think strategically, while the vehicles do the boring tasks of working through the detailed plans and scripts,” says Williams.
“Rich [Camilli, WHOI] and I are excited that we have made significant progress towards controlling a single autonomous vehicle robustly. This enables us to turn our focus to the vision that many have had, of allowing scientists to control larger networks of autonomous vehicles of many different types. In the future we would like to operate these networks persistently, without the current reliance on expensive ships and similar infrastructure.”
Professor Brian Williams, of aeronautics and astronautics at MIT
This research was funded in part by Schmidt Ocean Sciences. The underlying technology was supported in part by Boeing Co., the Keck Institute of Space Sciences, the Defense Advanced Research Projects Agency, and NASA.
Correction: several of the images used in this story were shared by researchers and should have been attributed to Schmidt Ocean Institute, a participant in the R/V Falkor expedition.
(As published in the June 2015 edition of Marine Technology Reporter - http://www.marinetechnologynews.com/Magazine)