determining the distribution of capabilities onboard
and on remote processors, as well as determining the
coordination of human and automation; and determining an autonomy deployment path that respects
best practices and ensures acceptance.
The results of the discussion in this section are
summarized in table 1. To complement and provide
detail to this discussion, we next describe a real-world
Case Study: Autonomy for sUAVs
In this section, we illustrate the conceptual design
process for autonomy with an example using sUAVs,
which have been proposed for a variety of commercial
(industrial and agricultural) and military applications,
including border interdiction, search and rescue,
wildfire suppression, communications relay, law en-
forcement, disaster and emergency management, and
three-dimensional archaeological map reconstruction.
A recent body of work has emerged in which autonomy capabilities have been proposed for sUAVs,
both fixed-wing and rotary wing. Designing autonomy for sUAVs is a different problem than designing
for larger platforms. More specifically, there is an
important trade-off between desired degrees of autonomy and the performance of the vehicle as captured by
its size, weight, and power requirements. Additionally,
sUAV platforms are typically cheap, modular, and easily
reconfigurable. This makes it possible to apply them to
highly specialized applications.
A number of technical challenges make developing
hardware and software systems for sUAVs more
difficult than for ground robots (Bachrach et al.
2011). Limited payload. This reduces the computational power available onboard and sometimes precludes the use of high-fidelity sensors.
Noisy position estimates. While sUAVs will generally have an inertial measurement unit (IMU),
double-integrating acceleration measurements from
lightweight microelectromechanical systems IMUs
results in large position errors.
Fast dynamics. sUAVs have fast and unstable dynamics, which result in a host of sensing, estimation, control, and planning challenges.
Constant motion. Unlike ground vehicles, an sUAV
cannot simply stop and perform more sensing or
computation when its state estimates have large
Planning in a three-dimensional representation of
an environment. A three-dimensional confi
guration space in general makes path planning
more computationally intensive.
Some design parameters that drive requirements for
sUAVs include whether the sUAV will be flown in an
indoor or an outdoor environment; whether daytime
or nighttime navigation is required, whether the
operating environment is GPS available or GPS denied, whether the environment is well mapped or
unknown, whether the environment is cluttered or
uncluttered, and whether the environment is con-
fined or open.
Another possible design consideration is cost. Lower-cost sUAVs — such as the AR.Drone, 2 in the price range
below $1500 — use less-expensive hardware for
onboard sensing and processing. Although they contain an IMU and often a GPS unit, measurement accuracy and stability are usually reduced. Consequently,
cameras and computer vision are often used for autonomous control to compensate for performance
limitations. On the other hand, higher-cost platforms, such as Asc Tec, offer improved flight stability
and more sophisticated sensing units, such as laser
rangefinders or thermal infrared cameras (Mathe and
Autonomous Search and Track
To illustrate the process of integrating autonomy into
the design of a complex robotic system, in this section
we consider the application of autonomy to a search
and track mission. In this application, an sUAV must
search for a target of interest, and once the target is
found it must be tracked until some terminal condition
is attained. For example, the target of interest might
be a human poaching rhino horns in South Africa
(Save the Rhino International 2015). The purpose of
this search and track application is to search for potential poachers; if one is detected, their location is
communicated to ground control; if the target is
moving, follow the target; terminate mission when
target is captured.
Autonomous search and track offers a useful illustration of the layered nature of autonomous behaviors.
Next, we examine the cognitive capabilities underlying
search and track and how they combine into an operational architecture.
Requirement Relevance to Autonomous Design
Mission purpose Cognitive components for accomplishing
mission goals and for safe operations
Operations Distributed architecture/human-machine
Constraints on operational architecture
Enabling versus enhancing technologies
Legalistic Human-machine architecture
Best practices P3I for integration of autonomy
Table 1. Summary of Requirements for
Conceptual Design and Relevance to Design for Autonomy.