24-08-2012, 02:36 PM
A Flight Control System for Aerial Robots: Algorithms and Experiments
[attachment=32635]
Abstract
This paper presents a hierarchical flight control system for unmanned aerial vehi-
cles. The proposed system executes high-level mission objectives by progressively
substantiating them into machine-level commands. The acquired information from
various sensors is propagated back to the higher layers for reactive decision mak-
ing. Each vehicle is connected via standardized wireless communication protocol
for scalable multi-agent coordination. The proposed system has been successfully
implemented on a number of small helicopters and validated in various applications.
Results from waypoint navigation, a probabilistic pursuit-evasion game and vision-
based target tracking demonstrate the potential of the proposed approach toward
intelligent flying robots.
INTRODUCTION
Deployment of intelligent robots has been made possible through technological
advances in various elds such as arti cial intelligence, robotics, wireless communication,
and control theories. There is little doubt that intelligent robots
will be employed to autonomously perform tasks, or embedded in many systems,
and extend our capabilities to perceive, reason and act, or substitute
human e orts in applications where human operation is dangerous, ine-
cient and/or impossible. Subscribing to this idea, BErkeley AeRobot (BEAR)
project aims to organize multiple autonomous agents into integrated and intelligent
systems with reduced cognition and control complexity, fault-tolerance.
FLIGHTMANAGEMENT SYSTEM FOR INTELLIGENT UN-
MANNED AERIAL VEHICLES
An \intelligent agent" continuously 1) perceives dynamically changing conditions
in its environment, 2) reasons to interpret perceived information, to
solve problems and to determine appropriate action, and 3) acts appropriately
to a ect conditions in its environment. Based on these attributes, this section
describes each layer in the hierarchical flight management system shown in
Fig. 2.
Sensing
Dynamically changing conditions in the environment and vehicle states are
perceived by various onboard sensors. Motion-related information, which is
vital for vehicle control and high-level operation, is measured by the onboard
navigation sensors such as inertial navigation system (INS) and global positioning
system (GPS). Additional sensors such as ultrasonic sensors and laser
range- nders are used to acquire the environment-speci c information including
relative distance from the ground surface, or to detect the objects in the
vicinity of the host vehicle. A computer vision system (Sharp et al. (2001)) is
used to detect objects of interest based on their color or shape.
Reasoning and Coordination
Fig. 2 shows three types of strategy planners to be implemented for each
experiment in Section 4. The appropriate strategy planner for a given mission
is selected by a switching layer.
When the current state of the world is not fully measurable, the world is modeled
as a partially observable Markov decision process (POMDP), as described
later in Section 4.2. The strategy planner then updates each agent's belief (in-
formation) state, i.e., probability distribution over the state space of the world,
given measurement and action histories, and generates a policy, i.e., a mapping
from the agent's belief state to its action set. Search of the optimal policy is
computationally intractable in most problems, thus usually sub-optimal policies
are implemented (Kim et al. (2001)), or, the class of policies to search
through is limited (Ng and Jordan (2000)). Algorithms are typically run on
real-time operating systems to satisfy hard real-time constraints.
The strategy planner also manages communication networks. Evolved from a
simple telemetry for data up/down link, the communication plays a vital role
in the real-time coordination and recon guration of multiple agents in dynamic
environment as a tightly coordinated, recon gurable, distributed networked
intelligence. Moreover, it is desirable to have the support of a high qualityof-
service (QoS) wireless communication system with minimal latency, in the
presence of ambient noise or signal jamming for secure operation.
Stabilization & Tracking Using Multi-loop Controller
Based on the identi ed model in Section 3.2, a stabilizing control law is designed.
In the rst approach, multiple single-input, single-output (SISO) control
loops are designed around the four inputs of longitudinal/lateral cyclic
pitches and main/tail collective pitches. This approach has obvious advantages
in terms of a simpler structure, straightforward design process, and low
computing load. On the other hand, it does not provide a systematic way to
account for uncertainty, disturbance, and saturation. Moreover, it has very
limited means to alleviate the coupling among channels.
The proposed controller consists of three loops: 1) innermost attitude controller,
2) mid-loop linear velocity controller, and 3) outer loop position controller
(Fig. 5).
The attitude controller feeds back only the deviation of the roll and pitch
angles from the trim condition (nonzero angle needed to maintain an equilibrium),
not the noisy angular rates p and q measured by rate gyros. This
approach yields a controller that is simpler and more robust to mechanical
vibration. The adequate angular feedback gains for roll and pitch channels are
determined to have acceptable response speed and damping ratio.
CONCLUSION
This paper presented a hierarchical RUAV flight control system. The vehicle
dynamics are identi ed as a linear model from the test flight data. The
tracking control layer is designed using the following two methods: multi-loop
PID control and nonlinear model predictive control. The performance of PID
controller has been validated in experiments that require a tracking trajectories
of moderate diculty. The nonlinear model predictive control has shown
an outstanding tracking performance in the presence of strong coupling and
control input saturation at the expense of heavier computation load. The proposed
multi-functional flight management system was tested in the following
examples: waypoint navigation, pursuit-evasion, tracking of a moving targets
and autonomous landing. Further research e ort will be made to expand the
capability of the flight management system with rich strategy planning logics,
increased robustness, and the wider flight envelope, hence narrowing down
the gap between current RUAVs and highly maneuverable flying robots with
intelligence.
[attachment=32635]
Abstract
This paper presents a hierarchical flight control system for unmanned aerial vehi-
cles. The proposed system executes high-level mission objectives by progressively
substantiating them into machine-level commands. The acquired information from
various sensors is propagated back to the higher layers for reactive decision mak-
ing. Each vehicle is connected via standardized wireless communication protocol
for scalable multi-agent coordination. The proposed system has been successfully
implemented on a number of small helicopters and validated in various applications.
Results from waypoint navigation, a probabilistic pursuit-evasion game and vision-
based target tracking demonstrate the potential of the proposed approach toward
intelligent flying robots.
INTRODUCTION
Deployment of intelligent robots has been made possible through technological
advances in various elds such as arti cial intelligence, robotics, wireless communication,
and control theories. There is little doubt that intelligent robots
will be employed to autonomously perform tasks, or embedded in many systems,
and extend our capabilities to perceive, reason and act, or substitute
human e orts in applications where human operation is dangerous, ine-
cient and/or impossible. Subscribing to this idea, BErkeley AeRobot (BEAR)
project aims to organize multiple autonomous agents into integrated and intelligent
systems with reduced cognition and control complexity, fault-tolerance.
FLIGHTMANAGEMENT SYSTEM FOR INTELLIGENT UN-
MANNED AERIAL VEHICLES
An \intelligent agent" continuously 1) perceives dynamically changing conditions
in its environment, 2) reasons to interpret perceived information, to
solve problems and to determine appropriate action, and 3) acts appropriately
to a ect conditions in its environment. Based on these attributes, this section
describes each layer in the hierarchical flight management system shown in
Fig. 2.
Sensing
Dynamically changing conditions in the environment and vehicle states are
perceived by various onboard sensors. Motion-related information, which is
vital for vehicle control and high-level operation, is measured by the onboard
navigation sensors such as inertial navigation system (INS) and global positioning
system (GPS). Additional sensors such as ultrasonic sensors and laser
range- nders are used to acquire the environment-speci c information including
relative distance from the ground surface, or to detect the objects in the
vicinity of the host vehicle. A computer vision system (Sharp et al. (2001)) is
used to detect objects of interest based on their color or shape.
Reasoning and Coordination
Fig. 2 shows three types of strategy planners to be implemented for each
experiment in Section 4. The appropriate strategy planner for a given mission
is selected by a switching layer.
When the current state of the world is not fully measurable, the world is modeled
as a partially observable Markov decision process (POMDP), as described
later in Section 4.2. The strategy planner then updates each agent's belief (in-
formation) state, i.e., probability distribution over the state space of the world,
given measurement and action histories, and generates a policy, i.e., a mapping
from the agent's belief state to its action set. Search of the optimal policy is
computationally intractable in most problems, thus usually sub-optimal policies
are implemented (Kim et al. (2001)), or, the class of policies to search
through is limited (Ng and Jordan (2000)). Algorithms are typically run on
real-time operating systems to satisfy hard real-time constraints.
The strategy planner also manages communication networks. Evolved from a
simple telemetry for data up/down link, the communication plays a vital role
in the real-time coordination and recon guration of multiple agents in dynamic
environment as a tightly coordinated, recon gurable, distributed networked
intelligence. Moreover, it is desirable to have the support of a high qualityof-
service (QoS) wireless communication system with minimal latency, in the
presence of ambient noise or signal jamming for secure operation.
Stabilization & Tracking Using Multi-loop Controller
Based on the identi ed model in Section 3.2, a stabilizing control law is designed.
In the rst approach, multiple single-input, single-output (SISO) control
loops are designed around the four inputs of longitudinal/lateral cyclic
pitches and main/tail collective pitches. This approach has obvious advantages
in terms of a simpler structure, straightforward design process, and low
computing load. On the other hand, it does not provide a systematic way to
account for uncertainty, disturbance, and saturation. Moreover, it has very
limited means to alleviate the coupling among channels.
The proposed controller consists of three loops: 1) innermost attitude controller,
2) mid-loop linear velocity controller, and 3) outer loop position controller
(Fig. 5).
The attitude controller feeds back only the deviation of the roll and pitch
angles from the trim condition (nonzero angle needed to maintain an equilibrium),
not the noisy angular rates p and q measured by rate gyros. This
approach yields a controller that is simpler and more robust to mechanical
vibration. The adequate angular feedback gains for roll and pitch channels are
determined to have acceptable response speed and damping ratio.
CONCLUSION
This paper presented a hierarchical RUAV flight control system. The vehicle
dynamics are identi ed as a linear model from the test flight data. The
tracking control layer is designed using the following two methods: multi-loop
PID control and nonlinear model predictive control. The performance of PID
controller has been validated in experiments that require a tracking trajectories
of moderate diculty. The nonlinear model predictive control has shown
an outstanding tracking performance in the presence of strong coupling and
control input saturation at the expense of heavier computation load. The proposed
multi-functional flight management system was tested in the following
examples: waypoint navigation, pursuit-evasion, tracking of a moving targets
and autonomous landing. Further research e ort will be made to expand the
capability of the flight management system with rich strategy planning logics,
increased robustness, and the wider flight envelope, hence narrowing down
the gap between current RUAVs and highly maneuverable flying robots with
intelligence.