US20150348235A1

US20150348235A1 - Distributed path planning for mobile sensors

Info

Publication number: US20150348235A1
Application number: US14/294,417
Authority: US
Inventors: Petros T. Boufounos; Richard Vaccaro; Mouhacine Benosman
Original assignee: Mitsubishi Electric Research Laboratories Inc
Current assignee: Mitsubishi Electric Research Laboratories Inc
Priority date: 2014-06-03
Filing date: 2014-06-03
Publication date: 2015-12-03
Also published as: JP6468941B2; JP2015230729A; US9218646B1

Abstract

A method plans paths of a set of mobile sensors with changeable positions and orientations in an environment. Each sensor includes a processor, an imaging system and a communication system. A desired resolution of coverage of the environment is defined, and an achieved resolution of the coverage is initialized. For each time instant and each sensor, an image of the environment is acquired using the imaging system. The achieved resolution is updated according to the image. The sensor is moved to a next position and orientation based on the achieved resolution and the desired resolution. Then, local information of the sensor is distributed to other sensors using the communication system to optimize a coverage of the environment.

Description

FIELD OF THE INVENTION

The invention relates generally to mobile sensors, and more particularly to distributed path planning for the sensors.

BACKGROUND OF THE INVENTION

Mobile sensor can be used to acquire images in a coordinated manner. The images can then be used in applications such as surveillance, cartography, and environmental monitoring. One problem in such systems is planning the paths the sensors should follow, a problem known as path planning.
In one such system with holonomic robots, where the controllable degrees of freedom are equal to the total degrees of freedom, anisotropic sensors with a bounded footprint are considered, see Hexsel et al., “Distributed Coverage Control for Mobile Anisotropic Sensor Networks,” Tech. Report CMU-RI-TR-13-01, Robotics Institute, Carnegie Mellon University, January 2013. That system models a 2-dimensional (2D) environment as a polygon, possibly containing obstacles. A fixed objective function maximizes a joint probability to detect objects. The objective function uses an a priori fixed density function that represents an importance of each point in the environment.

SUMMARY OF THE INVENTION

The embodiments of the invention provide a method for planning paths of a set of mobile sensors, that can be, for example, airborne, ground-based, or underwater. Each sensor includes an imaging system for imaging an environment. The imaging system can use optical imaging, synthetic aperture radar (SAR), hyperspectral imaging, physical aperture radar imaging, for example. Images acquired by the sensors can be used in surveillance, cartography and monitoring applications, among others.
At any time instant, each sensor moves to a next position and orientation to optimize a resolution of the imaging system is used to select a resolution for optimal coverage of the environment, according to a pre-specified desired resolution at each point in the environment. As used herein, the resolution depends on the size and density of pixels in the images. It should be noted that the coverage can be complete, or partial and images can overlap or not.
In a distributed manner, the motion of each sensor and its orientation are optimized by minimizing a cost function that characterizes how well the environment has been imaged, compared to the pre-specified resolution. To perform this optimization, each sensor communicates with neighboring sensors, and exchanges local information. The images acquired by the sensors can be combined into an overall image of the environment to achieve the desired resolution.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic of an example airborne mobile sensor system according to embodiments of the invention;

FIG. 2 is a side view of a sensor and environment in sensor coordinates according to embodiments of the invention;

FIG. 3 is a table giving variables, description, and exemplar values used by a model according to embodiments of the invention;

FIG. 4A is a schematic of an example environment according to embodiments of the invention;

FIG. 4B is a schematic of an irregular example environment;

FIG. 5 is a top view of the sensor footprint in sensor coordinates according to embodiments of the invention; and

FIG. 6 is a flow diagram of a method for planning paths of the sensors shown in FIG. 1 according to embodiments of the invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

FIG. 1 shows a set of mobile sensors 100 according to embodiments of our invention. The sensors can be airborne, ground-based or underwater, among others. The sensors can, for example, be arranged in indoor or outdoor drones, aircraft, or satellites. Each sensor includes a processor 101, an imaging system 102, and a communication system 103. The imaging system has a known footprint for imaging an environment 400, which may depend on the orientation of the sensor. The footprint is a projection of the imaging plane, such as a camera image plane or a radar beam pattern, onto the environment. The imaging system can use, among others, optical imaging, synthetic aperture radar (SAR), hyperspectral imaging, physical aperture radar imaging. For this reason, the term “image” is used broadly.
The sensors move along paths 102 to image the environment 400. The communication system includes a transceiver so that the sensors can communicate with each other over channels 105. In the preferred embodiment the channels are wireless using, e.g., radio or optical signals. By exchanging information between neighboring sensors, i.e., sensors within communication range, the sensors can perform the path planning in a distributed manner. However, it should be understood that the other communication techniques can also be used, for example some or all of the sensors can use digital fine wire tethers.
It is an objective to provide an image resolution over the environment that achieves a specified value at each point in the environment. As used herein, the resolution depends on the size and density of pixels in the images. The model uses a subadditive function to model the resolution provided by overlapping footprints 104. The model also uses an objective function that is time varying. At each time j, the objective function provides a measure of a difference between a desired resolution and a resolution achieved up to a previous time j−1.
Sensor Model
FIG. 2 is a side view of a sensor and environment in sensor coordinates. A point z in the environment is located on a line bisecting an angle γ_v. The table in FIG. 3 gives the variables 301, description 302 and exemplar values 303 used by our model. The variables include a height 311 of the sensor, horizontal 312 and vertical 313 angular widths, position 314 of the sensor, and declination 315 and azimuth 317 angles. The angles specify the orientation of the sensors. In the described embodiment, there are two degrees of freedom, however three degrees are not precluded.
For convenience, most of the subsequent computations use the angle ψ 316 to measure the declination, which is related to the actual declination angle φ using ψ=90°−φ. The labeled ranges in FIG. 2 are given by the following:
$\begin{matrix} Z_{\max} = H \tan (ψ + γ_{v} / 2) Z_{\min} = H \tan (ψ - γ_{v} / 2) z = H \tan (ψ) . & (1) \end{matrix}$
As shown in FIG. 4A, an example regular environment is a 100×100 polygonal region Q in the xy plane. An arbitrary point 401 in the environment is labeled qεQ. In sensor coordinates, the environment is the zy plane, which is a translated and rotated version of the x plane. A given sensor is located at a height H above the origin of the zv plane and the angle ψ is measured with respect to the z axis. The height H, and the x, y location specify the position of the sensors. The angles φ and ψ specify the orientation.
FIG. 4B shows an irregular example irregular environment 500.
Each sensor has the declination angle φ, with ψ=90°−φ. When the azimuth angle θ=0, the z-axis of the sensor is aligned with the x-axis of a global coordinate system. All variables associated with the i^thsensor have a superscript i for indexing the sensors.
FIG. 5 is an example of a top view of the sensor footprint 104 in sensor coordinates specific to an optical sensor. The footprint extends from Z_minto Z_maxas shown in FIG. 2. The footprint is a polygon defined by the four labeled vertices (1, 2, 3, 4). In sensor coordinates, the footprint vertices
(z_k, y_k) corresponding to the k^thvertex are
$\begin{matrix} [\begin{matrix} z_{1} & y_{1} \\ z_{2} & y_{2} \\ z_{3} & y_{3} \\ z_{4} & y_{4} \end{matrix}] = [\begin{matrix} Z_{\min} & - Z_{\min} \sin (γ h / 2) \\ Z_{\max} & - Z_{\max} \sin (γ h / 2) \\ Z_{\max} & Z_{\max} \sin (γ h / 2) \\ Z_{\min} & Z_{\min} \sin (γ h / 2) \end{matrix}] . & (2) \end{matrix}$
Let S(θ) be the rotation matrix defined below
$\begin{matrix} S (θ) = [\begin{matrix} \cos (θ) & - \sin (θ) \\ \sin (θ) & \cos (θ) \end{matrix}] . & (3) \end{matrix}$
The global coordinates (x, y) of a point (z, y) in the sensor footprint are obtained by rotating the point by the azumith angle θ, and translating the footprint by the camera location (c_x, c_y):
$\begin{matrix} [\begin{matrix} x \\ y \end{matrix}] = S (θ) [\begin{matrix} z \\ y \end{matrix}] + [\begin{matrix} c_{x} \\ c_{y} \end{matrix}] . & (4) \end{matrix}$
The four vertices in equation (2) defining the sensor footprint 104 can be transformed into global coordinates using equation (4). The sensor footprint is defined by four variables (c_x, c_y, θ, φ). The first two variables are the projection of the sensor position onto the environment plane, and the last two parameters are the horizontal and vertical angular variables. Other sensors may have different footprint shapes, with the footprint shape and size depending on the sensor orientation.
In most practical embodiments, the position parameters are updated on a relatively slow time scale because these parameters correspond to the physical position of the sensor, while the angular variables are updated on a relatively fast time scale because the angles can quickly change values. However, in some embodiments, position parameters and angle parameters might change values at the same time scale, or angle parameters might change values at a slower time scale.
Subadditive Combination of Overlapping Sensors
Assume that sensor i provides a resolution r_iin the footprint F_i, i=1, . . . , n. The problem is to model the resolution obtained in an intersection of complete or partial overlapping footprints. The best, and unrealistically optimistic, situation is for the overall resolution to be the sum of the individual resolutions. The worst, and unrealistically pessimistic, situation is for the overall resolution to equal the maximum of the individual sensor resolutions. The actual overall resolution is somewhere between these extremes.
That is, if
r=[r ₁ r ₂ . . . r _N] (5)
is a vector of the resolutions achieved by N sensors, the overall resolution res(r) obtained at points in the intersection of the sensor footprints satisfies the following inequalities
$\begin{matrix} \max_{i} r_{i} \leq res (r) \leq r_{1} + \dots + r_{N} . & (6) \end{matrix}$
One example of a function that satisfies this property is the l_pnorm of the vector r, 1<p<∞,
$\begin{matrix} { r }_{p} \overset{def}{=} {(r_{l}^{p} + \dots r_{N}^{p})}^{1 / p}, & (7) \end{matrix}$
where 1<p<∞. When p=1, the l_pnorm equals the upper bound in equation (6). When p=∞, the l_pnorm equals the lower bound in equation (6). Thus, a particular example of a subadditive model for the resolution obtained by overlapping sensors is the l_pnorm of a vector of individual resolutions, where 1<p<∞. Other embodiments can use different subadditive functions to model how images of different resolutions are combined.
Objective Function and Optimization
Let φ_d(q) be the desired resolution defined at every point q 401 in the environment 400. Let x_jbe a vector of the position variables of all of the sensors at time j. Let ψ_jand θ_jbe vectors corresponding to the vertical (declination) and horizontal (azimuth) angular variables at time j, respectively, of all of the sensors. Let Rⁱbe the resolution provided by the i^thsensor at all points in its footprint F_i, which is defined by sensor variables (cxⁱ,cyⁱ,θⁱ,ψⁱ) as
$\begin{matrix} R^{i} ({cx}^{i}, {cy}^{i}, θ^{i}, ψ^{i}, q) = {\begin{matrix} \frac{K}{H^{2} [1 + \tan^{2} (ψ^{i})]}, & q \in F_{i} ({cx}^{i}, {cy}^{i}, θ^{i}, ψ^{i}) \\ 0, & otherwise \end{matrix}, & (8) \end{matrix}$
where K is a sensor constant that depends on the number of pixels in an acquired image. If all of the sensors have the same value of K, then the value is unimportant for the optimization described below.
At any time j, the objective function we minimize is a difference between the desired resolution and an achieved resolution up to time j−1 according to the following function:
$\begin{matrix} G_{j} (x, θ, ψ) = \int_{Q} f (ϕ_{d} (q) - {[ϕ_{j - 1}^{P} (q) + \sum_{i} {(R^{i} ({cx}^{i} {cy}^{i}, θ^{i}, ψ^{i}, q))}^{p}]}^{1 / p}) \partial q, & (9) \end{matrix}$
where φ_j-1(q) is the resolution achieved by the sensors up to time j−1, p defines the norm used to model a subadditive combination of overlapping footprints, and ƒ(x) is a penalty function that penalizes deviation from the desired resolution.
For example, in one embodiment, ƒ(x)=x². This penalty function penalizes the achieved resolution when the resolution is lower or greater than the desired resolution. This forces the sensors to move to a different area of the environment when some of the sensors have been mapped to a sufficient resolution.
In another embodiment,
$f (x) = {\begin{matrix} x^{2}, & x \geq 0 \\ 0, & x < 0 \end{matrix} .$
This penalty function penalizes the achieved resolution only when it has not attained the desired resolution, which enables the sensor to continue improving the resolution of the imaged area beyond the pre-specified desired resolution. Of course, other embodiments may use other penalty functions.
Path Planning
FIG. 6 shows a method for path planning according to embodiments of the invention. By definition, the initial achieved resolution φ₀(q) is identically zero.
A gradient-based optimization is described by the following initialization and iterative steps. At each time j, a complete gradient-based minimization with respect to the angle parameters of the sensors is performed. However, sensor positions are updated using only a single gradient step. The reason is that after the sensors have moved and acquired new data, the objective function has changed.
Initialization
Given the desired resolution φ_d(q), and a vector x₀of initial sensor positions, initial sensor angles are determined 605 by the following optimization 611
$\begin{matrix} θ_{0}, ψ_{0} = \underset{θ, ψ}{\arg \min} G_{0} (x_{0} θ, ψ) . & (11) \end{matrix}$
The initial position gradient g₀is the gradient with respect to x of G₀(x, θ₀, ψ₀) evaluated at x₀.
Iteration 650 j=1, 2, . . . at Each Sensor for Each Time Step
Acquire 610 images 601 from all sensors 100, and update 620 an achieved resolution 621 according to
$\begin{matrix} ϕ_{j} (q, x_{j - 1}, θ_{j - 1}, ψ_{j - 1}) = {[ϕ_{j - 1}^{P} (q) + (\sum_{i} {(R^{i} ({cx}_{j - 1}^{i}, {cy}_{j - 1}^{i}, θ_{j - 1}^{i}, ψ_{j - 1}^{i}, q))}^{P})]}^{1 / p} . & (12) \end{matrix}$
Moving 630 the set of sensors to a next position and orientation of the sensor based on the achieved resolution and the desired resolution. The moving is in a direction of the negative position gradient 631
x _j =x _j-1 −αg _j-1 (13)
where α is a step size, and g_j-1is the position gradient at a previous time evaluated at position x_j-1. It should be understood that the moving to the next position and orientation can be a “null” move, i.e., the sensor remains in place.
Update 640 the sensor angular parameters and the position gradient 641 according to
$\begin{matrix} θ_{j}, ψ_{j} = \underset{θ, ψ}{\arg \min} G_{j} (x_{j}, θ, ψ) g_{j} = gradient with respect to x of G_{j} (x, θ_{j}, ψ_{j}) evaluated at x_{j}, & (14) \end{matrix}$
and iterate for the next time instant j=j+1.
Although the invention has been described by way of examples of preferred embodiments, it is to be understood that various other adaptations and modifications can be made within the spirit and scope of the invention. Therefore, it is the object of the appended claims to cover all such variations and modifications as come within the true spirit and scope of the invention.

Claims

We claim:

1. A method for planning paths of a set of sensors in an environment, wherein the sensors are mobile sensors with changeable positions and orientations, and each sensor includes a processor, an imaging system and a communication system, comprising the steps of:

defining a desired resolution of coverage of the environment;

initializing an achieved resolution of the coverage, and further comprising for each time instant in each processor of each sensor the steps of:

acquiring an image of the environment using the imaging system;

updating the achieved resolution according to the image; and

moving to a next position and orientation based on the achieved resolution and the desired resolution; and

distributing local information of the sensor to other sensors in the set of sensors using the communication system to optimize a coverage of the environment.

2. The method of claim 1, wherein the imaging system uses optical imaging.

3. The method of claim 1, wherein the imaging system uses synthetic aperture radar.

4. The method of claim 1, wherein the imaging system uses hyperspectral imaging.

5. The method of claim 1, wherein the imaging system uses physical aperture radar imaging.

6. The method of claim 1, wherein information comprises of the position, the orientation and the achieved resolution.

8. The method of claim 1, further comprising:

arranging the sensor in an indoor drone, an outdoor drone, an airplane, a satellite, a ground vehicle, a robot, or an underwater vehicle.

9. The method of claim 1, wherein the communication system includes wireless channels.

10. The method of claim 1, wherein the positions are updated at a different rate than the orientations.

11. The method of claim 1, wherein a model of the resolution is subadditive.

12. The method of claim 1, further comprising:

minimizing a difference between the desired resolution and the achieved resolution using an objective function.

13. The method of claim 12, wherein the objective function uses a function that penalizes a deviation from the desired resolution.

14. The method of claim 13, wherein the function penalizes the achieved resolution when the resolution is lower or greater than the desired resolution.

15. The method of claim 13, wherein the function penalizes the achieved resolution only when the achieved resolution has not attained the desired resolution.

16. The method of claim 1, wherein the moving is in a direction of a negative position gradient.