Cooperative AUV Navigation using a Single Surface Craft
Maurice F.Fallon,Georgios Papadopoulos and John J.Leonard英语四级吧
Abstract Maintaining accurate localization of an autonomous underwater vehicle (AUV)is difficult becau electronic signals such as GPS are highly attenuated by water making established land-bad localization systems,such as GPS,uless underwater.Instead we propo an alternative approach which integrates position information of other vehicles to reduce the error and uncertainty of the on-board position estimates of the AUV.This approach us the WHOI Acoustic Modem to exchange vehicle localization estimates—albeit at low transmission rates—while simultaneously estimating inter-vehicle range.The performance capabilities of the system were tested using Oceanrver’s Iver2and the MIT Scout kayaks.
slik>pasrby是什么意思啊
1Introduction
Localization or navigation of vehicles using only onboard local nsors,such as a Doppler Velocity Logger(DVL)or Inertial Measurement Unit(IMU),are certain to experience accumulated positioning error.One can,of cour,utilize more preci nsors to reduce the rate of accumulated error—DVL units with error accumula-tion rates as low as0.2%are commercially available.However this approac
h may not be satisfactory due to practical,power orfinancial limitations.
Regardless of the platform ud,the accumulation of error and uncertainty is simply slowed,rather than bounded.The result of this is that an AUV surveying the oceanfloor or a land robot building a street map must be halted on occasion so as to ret the position uncertainty—either by surfacing for a GPSfix or by repositioning at a known location.This procedure wastes both energy and time,requires a human interface and may be unacceptable in many operating environments.
The standard approach for bounding error underwater is Long Baline(LBL). Two or more beacons are deployed at known locations—either as buoys on the
2Maurice F.Fallon,Georgios Papadopoulos and John J.Leonard water surface or moored on the abed.The AUV transmits an acoustic query to the beacons which reply in a manner which allows the AUV to estimate the bea-con/AUV range and to then improve its own position estimate.Recent improve-ments to this system have removed the need for round-trip timing(Synced LBL) and also allowed for estimation of both range and angle using an array of receiving nsors(USBL).
While the technologies are now all commercially available,the mobility of the AUV is restricted as typical coverage is limited to an area within a few kilometers of the beacon.To relax this restriction an
alternative approach considers a system in which a surface vehicle(with access to GPS)or a submerged vehicle(with accurate dead reckoning instrumentation)communicates with afleet of much less accurately localized vehicles so as to improve the positioning of the latter.One example of this approach is the Moving Long Ba Line(MLBL)navigation propod by Vaganay et al.[9],in which typically two surface vehicles rve as mobile beacons for one or more AUVs.Other related recent rearch has been performed by Bahr et al.[1], Eustice et al.[4]and Maczka et al.[5].It should also be recognized that multi-AUV navigation falls within the wider problem of multi-robot cooperative localization, e[6]for a more general introduction to thefield.
In this paper,we describe experiments that extend the MLBL approach to situ-ations in which a single surface vehicle is ud to estimate the position of a sub-merged AUV using range-only measurements.In Section2the basic framework of this technique is discusd.Our algorithm is outlined in Section3followed by a number of modifications which improve performance.Section4prents the results of a combination of simulation and realistic experiments to illustrate the concept. Finally conclusions drawn from the experiments and the directions of future work are prented in Section5.default什么意思
2Cooperative Localization Under Water
This paper retains the framework for underwater localization previously introduced in[1]and also ud in[4].We shall assume there to be one surface vehicle providing the submergedfleet of vehicles with position information while perhaps operating as a communications moderator—in the dual role of a Communications and Nav-igation Aid(CNA).Each of the autonomous underwater vehicles maintains a dead reckoningfilter,drawing upon measurements of velocity,heading and depth.Fi-nally,communication through the water channel is possible using the WHOI Acous-tic Modem—at transmission rates of the order of32bytes per10conds—in a process which also yields a time-of-flight measurement which can be ud to esti-mate the inter-vehicle range.
牛津词典在线There are a number of methods which could be ud to integrate the received position information.Our earlier work,[1],propod an algorithm which utilized the on-board dead reckoning estimate of the AUV and a pair of CNA range estimates to form a complete estimate of the AUV state vector.
Cooperative AUV Navigation using a Single Surface Craft3 The abed,the water surface and deep a thermoclines within the water body have the ability to cau significant multi-path signal interference and the receipt of a substantial amount of infeasible outlier measurements.A typical datat was illustrated for a regular Long Baline systems in[8].For the reasons it would be reaso
nable to assume that the received measurement t obtained from the WHOI modem would contain substantial multi-modality,thereby motivating this approach.
However the advanced processing within the WHOI modem decoder has the abil-ity to suppress the bulk of the effects,such that the received measurements de-coded by the modem contain only a moderate amount of noi.For this reason the propod approach instead us an implementation of the Extended Kalman Filter.
健康早餐英语作文A particlefiltering approach[3]could also have been considered as this would have more accurately incorporated the non-linearity of the correction step,however becau we will maintain full control of the CNA’s motion this issue can broadly be avoided.
Previous proof-of-concept experiments illustrated that the range variance is broadly independent of range itlf,however detailed examination of this was not carried out [2].The modem transducer was then directly clamped to the underside of the kayak. Our more recent experiments have instead hung the transducer2-3meters below the kayak hull.We expect less noi interference from the kayak motor and less reflections from the water surface in this configuration.
Figure1illustrates WHOI modem range data plotted versus GPS-derived‘ground truth’,as measured i
n the Charles River adjacent to MIT recently.Becau the ground truth distance between the two vehicles was determined using impreci GPS measurements,it is difficult to precily estimate the distribution of the range measurements.Other issues,such as the position of the GPS nsor relative to the modem on the kayak must also be recognized.In the abnce of preci ground truth,we estimate the range variance to be between4–8m.科罗拉多州立大学
3Single Surface Craft Cooperative Navigation
magazine是什么意思The configuration we will consider in this work will be of a single CNA supporting N underwater vehicles1.Each AUV will maintain an estimate of its own position and uncertainty.This estimate will be propagated using the usual Kalman prediction step so as to integrate heading,forward and starboard velocity measurements.
As mentioned above,this estimate will be corrected using range and position in-formation relative to an CNA using the WHOI acoustic modem.At prent the32 byte packet transmitted from the CNA shall contain latitude,longitude,depth and heading as well as a UNIX time-stamp.Transmission of a packet consists of two stages:first a mini packet is transmitted to initiate the communication quence. The inter-vehicle range can be estimated using this mini packet.Following this,the information packet is transmitted in a process which lasts approximately5-6c-
expert什么意思4Maurice F.Fallon,Georgios Papadopoulos and John J.Leonard onds.In all,it is prudent to rerve10conds per transmission.Simularly the AUV will transmit a message containing its own position estimate as well the associated covariance matrix which can be ud to help the CNA plan its own supporting mo-tion—also requiring10conds per transmission.
snacksIt is envisaged that the MLBL will be integrated within a multi-AUV tup in which u of the communication channel is shared between many communicating process.As a result the transmission rate of a position/range pair is likely to be substantially below one measurement per10conds.Furthermore only a portion of transmitted messages will actually be received.For the reasons it is prudent to optimize the location from which the ASC transmits so as to maximize the ben-efit achieved from the correction step.Although a basic zig-zag motion plan was adopted in this work,future work will consider more elaborate motion planning for the CNA.
Fig.1Analysis of range estimates derived from the WHOI Modem.Upper Left:Comparison of modem range estimate(red dots)and range derived from GPS‘ground truth’(blue cross) for each fully successful10cond transmission period.Lower Left:Illustration of the frequency of successful transmissions.Category0reprents an entirely failed transmission;Category1: successful range transmission;Category2:successful range and packet transmission.Category 2corresponds to the modem ranges in upper left plot.Right:Histogram of range error(using estimated range versus GPS‘ground truth’range),also illustrated is a normal distributionfitted to the data(red,¯r=0.66m,σr=7.5m)and the normal distribution ud in the experiments in Section 4with(cyan,¯r=0m,σr=5m).This range data corresponds to Experiment1.
Cooperative AUV Navigation using a Single Surface Craft5 3.1Utilizing Partial Messages
As illustrated in Figure1,a significant proportion of the(range)mini packets are received without the information packet—meaning that the usual correction step cannot be made2
By linearly predicting the CNA position using previous position estimates,an estimate of the CNA at this time can be formed.This estimate can then be ud with the previously orphaned range measurement to allow another correction step to occur.While post processing of the data from the ex
periments prented in Section 4in this manner reduced the average error by approximately one meter,in future we propo to introduce redundancy into the transmitted messages so as to avoid this scenario.See Section5for more discussion.
3.2Online Compass Bias Correction
A Bayesianfilter-such as a Kalmanfilter or particlefilter-assumes that mea-surements are formed using unbiad estimators.Heading is however particularly difficult measurement to estimate properly.Compass accuracy can be effected by the characteristics of the local region,the magnetism of the vehicle itlf and mag-netic declination.It is particularly vere for impreci nsors ud aboard the CNA platform.As a result,the compass ud in the experiments prented in Section4 is a dominant source of navigation error.Typically compass bias is corrected using a calibration process which can be both complex and time consuming.In this sce-nario,the EKF corrections garnered using the CNA range and position can be ud to estimate the compass bias and to remove its effect.
Between successive corrections of the EKF,thefilter will be predicted according to the dynamical model.The frequency of the prediction step will be much higher than the correction step.The distance
between the posterior estimate of a correction step at time k1and the predicted position at time k2is the estimated relative
distance
Fig.2Compass Bias Correction Example:MLBL position estimate(blue)is corrected towards the ground truth(red)in a consistent direction.The angular correction of the4correction steps,θ1:4,can be ud to form an estimate of the bias angle,which is then removed.Note that multiple iterations of the prediction step take place between each correction step.
