Real-time motion planning for autonomous underwater vehicles

Real-time motion planning for autonomous underwater vehicles

REAL-TIME MOTTONPLANNTNG FOR AUTONOMOUS UNDERWATE ... 14th World Congress ofTFAC B-ld-02-5 Copyright © 1999 IFAC 14th Triennial World Congress, Bei...

4MB Sizes 0 Downloads 46 Views

REAL-TIME MOTTONPLANNTNG FOR AUTONOMOUS UNDERWATE ...

14th World Congress ofTFAC

B-ld-02-5

Copyright © 1999 IFAC 14th Triennial World Congress, Beijing, P.R. China

REAL-TIME MOTION PLANNING FOR AUTONOMOUS UNDERWATER VEHICLES Gianluca Antonelli ~ Stefano Chiaverini **

* PRl5MA Lab Dipartimento di Informatica e Sistemistica Universita degli Studi di Napoli Federico II Via Claudio 21, 80125 Napoli, Italy

[email protected] http://disna.dis.unina.it/robotics

** Dipartimento di Automazione, Elettromagnetismo, Ingegne1'ia dell'InJormazione e M atematica Industriale Universita degli Studi di Cassino Via G. Di Biasio 43, 03043 Cassino (FR) , Italy [email protected]

Abstract: In this paper a real-time navigation system for Autonomous Underwater Vehicles (AUVs) is presented. The navigation system is designed to allow fulfillment of a completely autonomous mission of AUVs in unknown and unstructured environments. Two reference missions are considered, namely pipeline/cable inspection and bottom exploration; their accomplishment requires the use of both heteroceptive and proprioceptive sensors. The motion planner provides on-line path generation with obstacle avoidance capabilities, while taking into account the non-holonomic motion constraints of the vehicle and the constant cruise requirement for proper sensor functioning. In addition, it must handle the effects of the ocean current on the vehicle dynamics. Extensive dynamic simulations of a vehicle in different unknown environments, subject to ocean current, have proven the effectiveness of the proposed navigation system. Copyright ~ 1999 IFAC Keywords: Autonomous vehicles, Trajectory planning, Navigation systems, Sensor fusion.

1. INTRODUCTION

Remotely-operated Underwater Vehicles (RUVs) constitute an important topic in the field of naval research. Among them, Autonomous Underwater Vehicles (AUVs) have received increasing attention since they allow to decrease mission costs and extend the vehicle autonomy.

These must be able to build environmental maps, extract relevant features of the environment from sensory data, and plan a safe motion trajectory. The development of a navigation system for AUVs is made very complex due to the following design

constraints: • it must work in real time so as to exploit sensory feedback from the environment; • the vehicle has a minimum turn radius giving non-holonomic motion constraints that must be accommodated in the trajectory planning;

To achieve autonomous missions through unknown, unstructured environments, underwater vehicles rely on advanced navigation systems. 545

Copyright 1999 IF AC

ISBN: 008 0432484

14th World Congress ofTFAC

REAL-TIME MOTION PLANNING FOR AUTONOMOUS UNDERWATE ...

• the environment influences the vehicle dynamics; • a constant cruise velocity of the vehicle might be required for proper functioning of some sensory system.

sensory data

Sensor Manager

Motion planning techniques can be divided in global and local ones. The former require a complete description of the environment in which the mission takes place; this is supposed to be \vell structured and accurately known in advance, thus usually leading to off-line implementations (see Latombe (1991) for an overview). For this reason, global techniques are typically unsuitable for application to autonomous vehicles moving in unstructured environments, e.g. , AUVs. On the other hand , Brooks (1986), Gat et al. (1994), Langer et al. (1994) developed local motion planners based on the behavior-control paradigm: single task modules realizing elementary behaviors are connected directly from sensors to actuators and work in parallel; composition of the elementary behaviors produces the resulting vehicle motion. As a different approach, several authors have used the so-called cell decomposition method (Borenstein and Koren (1994), Hyland and Taylor (1993), Oriolo et al. (1995)) according to which the environmental map is decomposed in cells; a cost is assigned to transit each cell and the best path is obtained via (either local or global) cost minimization. Finally, a popular local technique for vehicle motion planning is based on the potential field concept: the vehicle is subjected to virtual forces that push it toward the target while repulsing it from the obstacles; the virtual forces are then transformed in motion commands that can take into account non-holonomicity of the vehicle (see Arkin (1990), Borenstein and Koren (1994), Khatib (1987), Singh et al. (1996) . A recent review of motion planners for AUVs can be found in Yuh et al. (1996), where different experimental vehicles are shown with their navigation systems. In this paper a real-time navigation system for pipeline/cable inspection and bottom exploration is developed. In the case of the inspection task, tracking of the pipeline/cable must be accomplished without a-priori knowledge of its position; in t.he case of bottom exploration, a reference path has to be followed while avoiding collision with unknown obstacles. In both cases, at each time instant, the motion target is not a point fixed in advance but is rather computed on-line based on feedback from the sensory system. Nonholonomicity and minimum turn radius of the vehicle are explicitly taken into account; further, the cruise speed is kept in a range both to guarantee vehicle maneuverability also in presence of ocean current, and to ensure proper readings from

Supervisor

.......

Trajectory Generator

motion commands

17

Fig. 1. Basic architecture of the Navigation System. Doppler sonars. The environmental map is updated on-line from sonar information.

2. DESCRIPTION OF THE NAVIGATION SYSTEM

The Navigation System is composed of a Sensor Manager, a Supervisor, and a Trajectory Generator; its basic architecture is shown in Figure 1. The Sensor Manager has the task to elaborate the sensory data to provide information about the vehicle and the environment. The Supervisor has the task to avoid the vehicle get trapped by particular configurations of the environment; in this case, an escape algorithm must be commanded to the Trajectory Generator. The Trajectory Generator has the task to set the motion reference of the vehicle. It is mainly based on a Virtual Force Field (VFF) approach (Borenstein and Koren (1989)) , which is suitably modified so as to provide a more reactive response of the vehicle in case of obstacles occurring along the advancing direction.

2.1 Sensor Manager

In order to accomplish autonomous missions the vehicle must be equipped with both heteroceptive and proprioceptive sensors to get the required information about its position and attitude, its linear and angular velocity, actuator and control surfaces state, battery level, bottom distance, pipeline rela.tive position, and environmental information (e.g., relative obstacle position or ocean current). All those data must be handled by the Sensor Manager, that has the task to provide 546

Copyright 1999 IFAC

ISBN: 008 0432484

REAL-TIME MOTTONPLANNTNG FOR AUTONOMOUS UNDERWATE ...

the other modules with suitably elaborated and formatted information. In the case of underwater vehicles, direct position measurement may not be feasible: GPS based measures are applicable only to surface vehicles; ultra-sound positioning is not operative in open sea; tracking sonars require a support ship and high infrastructure cost; Long Baseline Acoustic Transponders provide poor performance measurements. In addition, precision in direct measurement systems of underwater vehicle position usually needs high costs or yields limited vehicle autonomy (Yoerger et al. (1996), Amat et al. (1996)). Integration measurement systems, instead, provide on board measurement at low cost; as a drawback, cumulative errors become significant for long duration missions. It should be noted that vehicle position measure-

ments are crucial in bottom survey of cables or pipelines; in this case, the errors on the bottom position are affected by the error on the absolute vehicle position. In fact, information about the environment surrounding the vehicle is more accurate than the absolute vehicle position. If the pipeline is provided with magnetic or acoustic reference via-points, the vehicle can reset the cumulative position error experienced by integration measurement systems. Vehicle velocity can be obtained by Doppler and/or electromagnetic based sensors (Amat et al. (1996»). This usually requires that the advancing velocity is constrained in a small range during the mission duration even in presence of ocean current. The environmental information acquisition and the path planning techniques are strongly affected by this requirement. Bottom distance and obstacles position are usually obtained by sonar sensors. At least two sonars are required: one vertical sonar providing the bottom distance, and one horizontal sonar, scanning an angular sector centered around the advancing direction, aimed at obstacle detection. The Sensor Manager must also take into account that in case of strong horizontal current the vehicle fore-aft direction is not coincident with the advancing direction. A solution to this problem is to adopt either several sonars that cover a wide angular sector or one programmable sonar that can be oriented to point the advancing direction. In order to build a map of the environment surrounding the vehicle, the Sensor Manager must update the past sensory data with the current ones so as to be properly expressed in the current vehicle-fixed frame. Storage of previous sensor readings obviously increase the information about the environment; however, a large map is accessed with low efficiency, while only local information is

14th World Congress ofTFAC

crucial to allow motion planning. Therefore, the Sensor Manager periodically purges unnecessary information. In case of exploration missions, the information that is purged from the map needed by the Trajectory Generator is stored to the purpose of off-line reconstruction of a global map of the mission. As for the pipeline inspection task, information about the pipeline position is also necessary. No off-line knowledge is fully reliable, since the ocean current may cause deplacement of the pipeline of several meters. Two methods can be used to measure the pipeline position: namely, visio-based and magnetic-based sensors (Kato et al. (1998)). With a visio-based system using more then one camera, position and attitude of the pipeline is available in all degrees of freedom. On the other hand, two magnetic sensors suitably located (in any case, far from magnetic noise sources, e.g., batteries) provide information on pipeline location as welL

2.2 Supervisor The Supervisor is the intelligence of the navigation system; its task is to prevent damage of the vehicle and to detect the occurrence of situations that can not be handled by the low level path planner (i.e., the Trajectory Generator). To the purpose, the Supervisor outputs a status label which instructs the Trajectory Generator about selection of proper algorithms to compute the motion commands to the vehicle; if a dangerous situation is detected, the vehicle status is changed and the Trajectory Generator activates a proper escape routine. The Supervisor has also the task to check the battery level, the elapsed mission time, and, in general, runs fault detection routines for sensor and actuators. Different approaches can be used to realize this module, e.g. finite-state systems, fuzzy logic, or expert systems. In Section 3 a Supervisor based on a finite-state system has been successfully tested; further research is ongoing to implement a fuzzylogic based technique. A typical problem with local motion planning techniques, such as the VFF approach used in the proposed Trajectory Generator, is the occurrence of local minima. These must properly be handled by the Supervisor. Another problem, affecting vehicles subject to non-holonomicmotion constraints (such as AUVs), is the possibility to get trapped in a canyonshaped environment. In fact, if the canyon happens to be closed, the vehicle might not revert its motion in a narrow space due to minimum turn radius constraints. 547

Copyright 1999 IF AC

ISBN: 008 0432484

REAL-TIME MOTTONPLANNTNG FOR AUTONOMOUS UNDERWATE ...

\

\

\

I "

\

I

I

\

\

~

\

'~

~

/

'--~~

14th World Congress ofTFAC

I

I

// /

I

,

/

/

4.:--

---~_I_~

Fig. 2. A sketch of the two turn semi-circles in the case of forward advancing without ocean current. As an example, of the capabilities of the Supervisor tested in Section 3, the strategy adopted to detect a canyon-shaped environment is briefly described. Two semi-circles with center in the origin of the vehicle-fixed frame and diameter parallel to the advancing direction are first defined; note that the two semi-circles are aligned with the fore-aft direction only if the vehicle is moving forward without ocean current. Figure 2 shows the defined semi-circles on the horizontal plane of the vehicle. Similar semi-circles can be defined in the vertical plane. The radius of the semi-circles is set as twice the minimum turn radius of the vehicle. It can be easily recognized that, to guarantee full maneuverability of the vehicle, obstacles cannot occur in both semi-circles at a time in view of turn radius and velocity constraints. Remarkably, the minimum turn radius depends on the magnitude and direction of the ocean current, because the control surfaces provide an action that is a function of the effective water-vehicle relative velocity. For this reason, the radius of the two semi-circles is computed on-line to reflect the actual vehicle maneuverability at each time instant. When both semi-circles are violated, the Supervisor changes the status label to signal t.he canyon trap. The corresponding escape routine implemented in the Trajectory Generator makes the vehicle to turn in the direction of t.he last semi-circle violated, until the integrity of the two semi-circles is reconstituted. Notice that this will happen at least after a 1800 turn.

2.3 Trajectory Generator Depending on the vehicle status and the environmental information, the Trajectory Generator has

Fig. 3. Avoiding an obst.acle occurring along the forward direction.

the task to generate the reference values of the low-level motion control system. Underwater vehicles used in pipeline/cable inspection are usually driven by one or more thrusters and by two control surfaces: the thrusters mainly provide motion in the advancing direction, while the control surfaces are used to vary yaw and pitch angles. Vehicle shape and wing location usually • provide stabilization of the roll angular motion. During the normal operating mode the desired path is obtained via a VFF approach, as the ODe presented in Borenstein and Koren (1989). In the following, a method to efficiently avoid obstacles in the forward direction is described. Figure 3 shows a definition of the semi-circles different from the above, in that they now overlap in the advancing direction of an angle o. An obstacle occurring along the advancing direction is seen by the Supervisor as a canyon-like trap, causing the Trajectory Generation to command a turn until the integrity of one of the two semicircle is reconstitut.ed. This means that the vehicle will turn until the obstacle is no longer in the angular sector of width 8; from that time on, it will be driven by the usual VFF path generation. Finally, the Trajectory Generator has the task t.o generate the current motion target. The mission can be seen as the task of tracking an unknown path; information on the path is obtained by magnetic or visio-based sensors, and the motion target must thus be generated on-line. The Trajectory Generation defines the current target by extending the pipeline trace along its currently estimated direction. The position and the direction of the pipeline can be provided by direct sensor measurement or, when the vehicle is far from it (e.g., because of an obstacle avoidance maneuver), from reconstructed data. In Figure 4 an example of current target generation is sketched. 548

Copyright 1999 IF AC

ISBN: 008 0432484

REAL-TIME MOTTONPLANNTNG FOR AUTONOMOUS UNDERWATE ...

14th World Congress ofTFAC

-50=----==---~----~----~----"

current • target

Fig. 4. Example of generation of the current target.

3. SIMULATION

50~==~=---~----~----~----~

40

100

o

50L---4~0------60L-----8~O-----1~OO-----1~2~O

Fig. 6. Obstacle in the forward direction (2of4) -SO.---~r---~------~----'---~~

In Figure 5, the vehicle detects the violation of the turn semi-circles and starts to turn until the integrity of one of them is reconstituted; in Figure 6, the Trajectory Generator commands to turn back to the target on the pipeline, that has been updated via an estimate of the actual position of the pipeline (since the pipeline is out of the vision system range); in Figure 7, the obstacles do not affect anymore the generation of the motion command, while in Figure 8 the vehicle sees again the pipeline and can update the target estimate. Finally, in Figure 9 part of a complete simulated mission is shown. The vehicle is approaching the pipeline while the horizontal sonar detects the presence of an obstacle in the forward direction. Remarkably, the vehicle sees the pipeline only in two points (at around 100 m and 300 m

80

Fig. 5. Obstacle in the forward direction (lof4)

Several dynamic simulations have been run under different environmental conditions. The vehicle shows a safe behavior with respect to changes in environmental conditions; due to an efficient management of sensor information it tracks the cable/pipeline while avoiding entering in narrow spaces. In the following, examples of the obtained results are reported. Figures 5 to 8 show four phases of an obstacle avoidance maneuver in the case of an obstacle occurring along the forward motion direction. The dash-dotted line is the trace of the pipeline/cable (or the desired path in case of an exploration mission), the solid line is the path followed by the vehicle, the symbol x indicates the presence of an obstacle, the symbol 0 represents the current target. The inner circle represents the safe-turn area (of radius twice the minimum turn radius of""" 15 m), while the outer circle shows the vehicle scene horizon (of radius equal to the horizontal sonar range of >=::: 50 m); this gives an idea of how is critical for the vehicle to generate a safe path.

60

50L---~----~----~----~----~~

40

60

80

100

120

Fig. 7. Obstacle in the forward direction (30f4) along x, where the pipeline is under the vision system); therefore, the current target during the whole obstacle avoidance maneuver is obtained by estimate from previous sensor information. 549

Copyright 1999 IF AC

ISBN: 008 0432484

REAL-TIME MOTTONPLANNTNG FOR AUTONOMOUS UNDERWATE ...

14th World Congress ofTFAC

5. BIBLIOGRAPHY Amat, .L, J. Battle, A. Casals, and J. Forest (1996). Garbi: A Low Cost ROV, Constraints and Solutions. Pre-Proc. of the 6th IARP on

Underwater Robotics. Arkin, RC. (1990). The Impact of Cybernetics on the Design of a Mobile Robot System: A Case Study. IEEE Trans. on Systems, Man and Cybernetics, 20(6), pp 1245-1257. Brooks, R.A. (1986). A Robust. Layered Control System for a Mobile Robot. IEEE J. of Robotics and Automation, 2(1), pp 14-23. Borenstein, J. and Y. Koren (1989). Real-Time Obstacle Avoidance for Fast Mobile Robots. IEEE Trans. on Systems, Man and Cybernetics, 19(5), pp 1179-1187. Gat, E., R. Desai, R. Ivlev, J. Loch, and D.P. Miller (1994). Behavior Control for Robotic Exploration of Planetary Surfaces. IEEE Trans. on Robotics and Automation, 10(4), pp 490503. Hyland, J.C. and F.J. Taylor (1993). Mine Avoidance Techniques for Underwater Vehicles. IEEE J. of Oceanic Engineering, 18(3), pp 34Q-350. Kato, N., J. Kojima, Y. Kato, S. :Matumoto, and K. Asakawa (1998). Optimization of Configuration of Autonomous Underwater Vehicle for Inspection of Underwater Cables. Proc. IEEE Int. Conf. on Robotics and Automation, pp 10451050. Khatib O. (1987). Real-Time Obstacle Avoidance for Manipulators and Mobile Robots. Proc. IEEE Int. Conf. on Robotics and Automation,

Fig. 8. Obstacle in the forward direction (40f4) xy plane

pp 500--505. 50

100

150 200 [m]

250

Langer, D., J.K. Rosenblatt, and M. Hebert (1994). A Behavior-Based System for Off-Road Navigation. IEEE Tmns. on Robotics and Automation, 10(6), pp 776-783. Latombe, J.C. (1991). Robot Motion Planning. Kluwer Academic Press, USA. Oriolo, G., M. Vendittelli, and G. Ulivi (1995). On-Line Map Building and Navigation for Autonomous Mobile Robots. Proc. IEEE Int. Conf. on Robotics and Automation, pp 2900-

300

Fig. 9. Planar view of a simulated motion. 4. CONCLUSIONS In this paper a real-time navigation system for Autonomous Underwater Vehicles has been presented. The developed navigation system allows fulfillment of a completely autonomous mission of an AUV in unknown and unstructured environments. Integration of heteroceptive and proprioceptive sensors is adopted to accomplish missions of pipeline/cable inspection and bottom exploration. The navigation system complies to nonholonomic motion constraints of the vehicle and provides on-line pat.h generation with obstacle avoidance capability. Dynamic simulations of a vehicle in several cases of unknown environments, subject to ocean current, have proven the effectiveness of the proposed navigation system technique. Examples of the obtained simulation results are reported in the paper.

2906.

Singh, L., H. Stephanou, and J. Wen (1996). Real-Time Robot Motion Control with Circulatory Fields. Proc. IEEE Int. Conf. on Robotics and Automation, pp 2737-2742. Yoerger, D.R., A.M. Bradley, B.B. Walden, H. Singh, and R Bachmayer (1996). Surveying a Subsea Lava Flow Using the Autonomous Benthic Explorer (ABE). Prc-Proc. of the 6th JARP on Unde'f"Water Robotics. Yuh, J., T. Ura, and G.A. Bekey (Eds.) (1996). Underwater Robots, Kluwer Academic Publishers, USA. 550

Copyright 1999 IF AC

ISBN: 008 0432484