Code for autonomous ground vehicle, Data Bus, 3rd place winner in 2012 Sparkfun AVC.
Dependencies: Watchdog mbed Schedule SimpleFilter LSM303DLM PinDetect DebounceIn Servo
updater.c
- Committer:
- shimniok
- Date:
- 2012-06-20
- Revision:
- 0:826c6171fc1b
File content as of revision 0:826c6171fc1b:
#include "Config.h" #include "Actuators.h" #include "Sensors.h" #include "SystemState.h" #include "GPS.h" #include "Steering.h" #include "Servo.h" #include "Mapping.h" #include "CartPosition.h" #include "GeoPosition.h" #include "kalman.h" #define _x_ 0 #define _y_ 1 #define _z_ 2 // Every ... // 10ms (100Hz) -- update gyro, encoders; update AHRS; update navigation; set log data // 10ms (100Hz) -- camera (actually runs 30fps // 20ms (50Hz) -- update compass sensor // 50ms (20Hz) -- create new log entry in log FIFO // 100ms (10Hz) -- GPS update // 100ms (10Hz) -- control update #define CTRL_SKIP 5 #define MAG_SKIP 2 #define LOG_SKIP 5 int control_count=CTRL_SKIP; int update_count=MAG_SKIP; // call Update_mag() every update_count calls to schedHandler() int log_count=LOG_SKIP; // buffer a new status entry for logging every log_count calls to schedHandler int tReal; // calculate real elapsed time // TODO: 3 better encapsulation, please extern Sensors sensors; extern SystemState state[SSBUF]; extern unsigned char inState; extern unsigned char outState; extern bool ssBufOverrun; extern GPS gps; extern Mapping mapper; extern Steering steerCalc; // steering calculator extern Timer timer; extern DigitalOut ahrsStatus; // AHRS status LED // Navigation extern Config config; int nextWaypoint = 0; // next waypoint destination int lastWaypoint = 1; double bearing; // bearing to next waypoint double distance; // distance to next waypoint float steerAngle; // steering angle float cte; // Cross Track error float cteI; // Integral of Cross Track Error // Throttle PID float speedDt=0; // dt for the speed PID float integral=0; // error integral for speed PID float lastError=0; // previous error, used for calculating derivative bool go=false; // initiate throttle (or not) float desiredSpeed; // speed set point float nowSpeed; // Pose Estimation bool initNav = true; float initHdg = true; // indiciates that heading needs to be initialized by gps float initialHeading=-999; // initial heading float myHeading=0; // heading sent to KF CartPosition cartHere; // position estimate, cartesian GeoPosition here; // position estimate, lat/lon extern GeoPosition gps_here; int timeZero=0; int lastTime=-1; // used to calculate dt for KF int thisTime; // used to calculate dt for KF float dt; // dt for the KF float lagHeading = 0; // lagged heading estimate //GeoPosition lagHere; // lagged position estimate; use here as the current position estimate float errAngle; // error between gyro hdg estimate and gps hdg estimate float gyroBias=0; // exponentially weighted moving average of gyro error float Ag = (2.0/(1000.0+1.0)); // Gyro bias filter alpha, gyro; # of 10ms steps float Kbias = 0.995; float filtErrRate = 0; float biasErrAngle = 0; #define MAXHIST 128 // must be multiple of 0x08 #define inc(x) (x) = ((x)+1)&(MAXHIST-1) struct { float x; // x coordinate float y; // y coordinate float hdg; // heading float dist; // distance float gyro; // heading rate float ghdg; // uncorrected gyro heading float dt; // delta time } history[MAXHIST]; // fifo for sensor data, position, heading, dt int hCount=0; // history counter; one > 100, we can go back in time to reference gyro history int now = 0; // fifo input index int prev = 0; // previous fifo iput index int lag = 0; // fifo output index int lagPrev = 0; // previous fifo output index /** set flag to initialize navigation at next schedHandler() call */ void restartNav() { initNav = true; initHdg = true; return; } /** instruct the controller to throttle up */ void beginRun() { go = true; timeZero = thisTime; // initialize return; } void endRun() { go = false; initNav = true; return; } /** get elasped time in update loop */ int getUpdateTime() { return tReal; } void setSpeed(float speed) { if (desiredSpeed != speed) { desiredSpeed = speed; integral = 0; } return; } /** schedHandler fires off the various routines at appropriate schedules * */ void update() { tReal = timer.read_us(); ahrsStatus = 0; thisTime = timer.read_ms(); dt = (lastTime < 0) ? 0 : ((float) thisTime - (float) lastTime) / 1000.0; // first pass let dt=0 lastTime = thisTime; // Add up dt to speedDt speedDt += dt; // Log Data Timestamp int timestamp = timer.read_ms(); ////////////////////////////////////////////////////////////////////////////// // NAVIGATION INIT ////////////////////////////////////////////////////////////////////////////// // initNav is set with call to restartNav() when the "go" button is pressed. Up // to 10ms later this function is called and the code below will initialize the // dead reckoning position and estimated position variables // if (initNav == true) { initNav = false; here.set(config.wpt[0]); nextWaypoint = 1; // Point to the next waypoint; 0th wpt is the starting point lastWaypoint = 1; // Init to waypoint 1, we we don't start from wpt 0 at the turning speed // Initialize lag estimates //lagHere.set( here ); // Initialize fifo hCount = 0; now = 0; // initialize what will become lag data in 1 second from now history[now].dt = 0; // initial position is waypoint 0 history[now].x = config.cwpt[0]._x; history[now].y = config.cwpt[0]._y; cartHere.set(history[now].x, history[now].y); // initialize heading to bearing between waypoint 0 and 1 //history[now].hdg = here.bearingTo(config.wpt[nextWaypoint]); history[now].hdg = cartHere.bearingTo(config.cwpt[nextWaypoint]); history[now].ghdg = history[now].hdg; initialHeading = history[now].hdg; // Initialize Kalman Filter headingKalmanInit(history[now].hdg); // initialize cross track error cte = 0; cteI = 0; // point next fifo input to slot 1, slot 0 occupied/initialized, now lag = 0; lagPrev = 0; prev = now; // point to the most recently entered data now = 1; // new input slot } ////////////////////////////////////////////////////////////////////////////// // SENSOR UPDATES ////////////////////////////////////////////////////////////////////////////// // TODO: 3 This really should run infrequently sensors.Read_Power(); sensors.Read_Encoders(); nowSpeed = sensors.encSpeed; sensors.Read_Gyro(); sensors.Read_Rangers(); //sensors.Read_Camera(); ////////////////////////////////////////////////////////////////////////////// // Obtain GPS data ////////////////////////////////////////////////////////////////////////////// // synchronize when RMC and GGA sentences received w/ AHRS // Do this in schedHandler?? GPS data is coming in not entirely in sync // with the logging info if (gps.nmea.rmc_ready() && gps.nmea.gga_ready()) { char gpsdate[32], gpstime[32]; gps.process(gps_here, gpsdate, gpstime); //gpsStatus = (gps.hdop > 0.0 && gps.hdop < 3.0) ? 1 : 0; // update system status struct for logging state[inState].gpsLatitude = gps_here.latitude(); state[inState].gpsLongitude = gps_here.longitude(); state[inState].gpsHDOP = gps.hdop; state[inState].gpsCourse = gps.nmea.f_course(); state[inState].gpsSpeed = gps.nmea.f_speed_mph(); // can we just do kph/1000 ? or mps? state[inState].gpsSats = gps.nmea.sat_count(); } ////////////////////////////////////////////////////////////////////////////// // HEADING AND POSITION UPDATE ////////////////////////////////////////////////////////////////////////////// // TODO: 2 Position filtering // position will be updated based on heading error from heading estimate // TODO: 2 Distance/speed filtering // this might be useful, but not sure it's worth the effort // So the big pain in the ass is that the GPS data coming in represents the // state of the system ~1s ago. Yes, a full second of lag despite running // at 10hz (or whatever). So if we try and fuse a lagged gps heading with a // relatively current gyro heading rate, the gyro is ignored and the heading // estimate lags reality // // In real life testing, the robot steering control was highly unstable with // too much gain (typical of any negative feedback system with a very high // phase shift and too much feedback). It'd drive around in circles trying to // hunt for the next waypoint. Dropping the gain restored stability but the // steering overshot significantly, because of the lag. It sort of worked but // it was ugly. So here's my lame attempt to fix all this. // // We'll find out how much error there is between gps heading and the integrated // gyro heading from a second ago. // stick precalculated gyro data, with bias correction, into fifo //history[now].gyro = sensors.gyro[_z_] - gyroBias; history[now].gyro = sensors.gyro[_z_]; // Have to store dt history too (feh) history[now].dt = dt; // Store distance travelled in a fifo for later use history[now].dist = (sensors.lrEncDistance + sensors.rrEncDistance) / 2.0; // Calc and store heading history[now].ghdg = history[prev].ghdg + dt*history[now].gyro; // raw gyro calculated heading //history[now].hdg = history[prev].ghdg - dt*gyroBias; // bias-corrected gyro heading history[now].hdg = history[prev].hdg + dt*history[now].gyro; if (history[now].hdg >= 360.0) history[now].hdg -= 360.0; if (history[now].hdg < 0) history[now].hdg += 360.0; //fprintf(stdout, "%d %d %.4f %.4f\n", now, prev, history[now].hdg, history[prev].hdg); // We can't do anything until the history buffer is full if (hCount < 100) { // Until the fifo is full, only keep track of current gyro heading hCount++; // after n iterations the fifo will be full } else { // Now that the buffer is full, we'll maintain a Kalman Filtered estimate that is // time-shifted to the past because the GPS output represents the system state from // the past. We'll also take our history of gyro readings from the most recent // filter update to the present time and update our current heading and position //////////////////////////////////////////////////////////////////////////////// // UPDATE LAGGED ESTIMATE // Recover data from 1 second ago which will be used to generate // updated lag estimates // GPS heading is unavailable from this particular GPS at speeds < 0.5 mph bool useGps = (state[inState].gpsLatitude != 0 && state[inState].lrEncSpeed > 1.0 && state[inState].rrEncSpeed > 1.0 && (thisTime-timeZero) > 3000); // gps hdg converges by 3-5 sec. // This represents the best estimate for heading... for one second ago // If we do nothing else, the robot will think it is located at a position 1 second behind // where it actually is. That means that any control feedback needs to have a longer time // constant than 1 sec or the robot will have unstable heading correctoin. // Use gps data when available, always use gyro data // Clamp heading to initial heading when we're not moving; hopefully this will // give the KF a head start figuring out how to deal with the gyro if (go) { lagHeading = headingKalman(history[lag].dt, state[inState].gpsCourse, useGps, history[lag].gyro, true); } else { lagHeading = headingKalman(history[lag].dt, initialHeading, true, history[lag].gyro, true); } // TODO 1: need to figure out exactly how to update t-1 position correctly // Update the lagged position estimate //lagHere.move(lagHeading, history[lag].dist); history[lag].x = history[lagPrev].x + history[lag].dist * sin(lagHeading); history[lag].y = history[lagPrev].y + history[lag].dist * cos(lagHeading); //////////////////////////////////////////////////////////////////////////////// // UPDATE CURRENT ESTIMATE // Now we need to re-calculate the current heading and position, starting with the most recent // heading estimate representing the heading 1 second ago. // // Nuance: lag and now have already been incremented so that works out beautifully for this loop // // Heading is easy. Original heading - estimated heading gives a tiny error. Add this to all the historical // heading data up to present. // // For position re-calculation, we iterate 100 times up to present record. Haversine is way, way too slow, // so is trig calcs is marginal. Update rate is 10ms and we can't hog more than maybe 2-3ms as the outer // loop has logging work to do. Rotating each point is faster; pre-calculate sin/cos, etc. // // initialize these once errAngle = (lagHeading - history[lag].hdg); if (errAngle <= -180.0) errAngle += 360.0; if (errAngle > 180) errAngle -= 360.0; //fprintf(stdout, "%d %.2f, %.2f, %.4f %.4f\n", lag, lagHeading, history[lag].hdg, lagHeading - history[lag].hdg, errAngle); float cosA = cos(errAngle * PI / 180); float sinA = sin(errAngle * PI / 180); // Start at the out side of the fifo which is from 1 second ago int i = lag; for (int j=0; j < 100; j++) { history[i].hdg += errAngle; // Rotate x, y by errAngle around pivot point; no need to rotate pivot point (j=0) if (j > 0) { float dx = history[i].x-history[lag].x; float dy = history[i].y-history[lag].y; history[i].x = history[lag].x + dx*cosA - dy*sinA; history[i].y = history[lag].y + dx*sinA + dy*cosA; } inc(i); } // gyro bias, correct only with shallow steering angles // if we're not going, assume heading rate is 0 and correct accordingly // If we are going, compare gyro-generated heading estimate with kalman // heading estimate and correct bias accordingly using PI controller with // fairly long time constant // TODO: 3 Add something in here to stop updating if the gps is unreliable; need to find out what indicates poor gps heading if (history[lag].dt > 0 && fabs(steerAngle) < 5.0 && useGps) { biasErrAngle = Kbias*biasErrAngle + (1-Kbias)*(history[lag].ghdg - state[inState].gpsCourse); // can use this to compute gyro bias if (biasErrAngle <= -180.0) biasErrAngle += 360.0; if (biasErrAngle > 180) biasErrAngle -= 360.0; float errRate = biasErrAngle / history[lag].dt; //if (!go) errRate = history[lag].gyro; // Filter bias based on errRate which is based on filtered biasErrAngle gyroBias = Kbias*gyroBias + (1-Kbias)*errRate; //fprintf(stdout, "%d %.2f, %.2f, %.4f %.4f %.4f\n", lag, lagHeading, history[lag].hdg, errAngle, errRate, gyroBias); } // make sure we update the lag heading with the new estimate history[lag].hdg = lagHeading; // increment lag pointer and wrap lagPrev = lag; inc(lag); } state[inState].estHeading = history[lag].hdg; // the variable "here" is the current position //here.move(history[now].hdg, history[now].dist); float r = PI/180 * history[now].hdg; // update current position history[now].x = history[prev].x + history[now].dist * sin(r); history[now].y = history[prev].y + history[now].dist * cos(r); cartHere.set(history[now].x, history[now].y); mapper.cartToGeo(cartHere, &here); // TODO: don't update gyro heading if speed ~0 -- or use this time to re-calc bias? // (sensors.lrEncSpeed == 0 && sensors.rrEncSpeed == 0) ////////////////////////////////////////////////////////////////////////////// // NAVIGATION UPDATE ////////////////////////////////////////////////////////////////////////////// //bearing = here.bearingTo(config.wpt[nextWaypoint]); bearing = cartHere.bearingTo(config.cwpt[nextWaypoint]); //distance = here.distanceTo(config.wpt[nextWaypoint]); distance = cartHere.distanceTo(config.cwpt[nextWaypoint]); //float prevDistance = here.distanceTo(config.wpt[lastWaypoint]); float prevDistance = cartHere.distanceTo(config.cwpt[lastWaypoint]); double relativeBrg = bearing - history[now].hdg; // if correction angle is < -180, express as negative degree // TODO: 3 turn this into a function if (relativeBrg < -180.0) relativeBrg += 360.0; if (relativeBrg > 180.0) relativeBrg -= 360.0; // if within config.waypointDist distance threshold move to next waypoint // TODO: 3 figure out how to output feedback on wpt arrival external to this function if (go) { // if we're within brakeDist of next or previous waypoint, run @ turn speed // This would normally mean we run at turn speed until we're brakeDist away // from waypoint 0, but we trick the algorithm by initializing prevWaypoint to waypoint 1 if (distance < config.brakeDist || prevDistance < config.brakeDist) { setSpeed( config.turnSpeed ); } else if ( (thisTime - timeZero) < 1000 ) { setSpeed( config.startSpeed ); } else { setSpeed( config.topSpeed ); } if (distance < config.waypointDist) { //fprintf(stdout, "Arrived at wpt %d\n", nextWaypoint); //speaker.beep(3000.0, 0.5); // non-blocking lastWaypoint = nextWaypoint; nextWaypoint++; cteI = 0; } } else { setSpeed( 0.0 ); } // Are we at the last waypoint? // currently handled external to this routine ////////////////////////////////////////////////////////////////////////////// // OBSTACLE DETECTION & AVOIDANCE ////////////////////////////////////////////////////////////////////////////// // TODO: 1 limit steering angle based on object detection ? // or limit relative brg perhaps? // TODO: 1 add vision obstacle detection and filtering // TODO: 1 add ranger obstacle detection and filtering/fusion with vision ////////////////////////////////////////////////////////////////////////////// // CONTROL UPDATE ////////////////////////////////////////////////////////////////////////////// if (--control_count == 0) { // Compute cross track error cte = steerCalc.crossTrack(history[now].x, history[now].y, config.cwpt[lastWaypoint]._x, config.cwpt[lastWaypoint]._y, config.cwpt[nextWaypoint]._x, config.cwpt[nextWaypoint]._y); cteI += cte; // Use either pure pursuit algorithm or original algorithm if (config.usePP) { steerAngle = steerCalc.purePursuitSA(state[inState].estHeading, history[now].x, history[now].y, config.cwpt[lastWaypoint]._x, config.cwpt[lastWaypoint]._y, config.cwpt[nextWaypoint]._x, config.cwpt[nextWaypoint]._y); } else { steerAngle = steerCalc.calcSA(relativeBrg, config.minRadius); // use the configured minimum turn radius } // Apply gain factor for near straight line if (fabs(steerAngle) < config.steerGainAngle) steerAngle *= config.steerGain; // Curb avoidance if (sensors.rightRanger < config.curbThreshold) { steerAngle -= config.curbGain * (config.curbThreshold - sensors.rightRanger); } setSteering( steerAngle ); // void throttleUpdate( float speed, float dt ) // PID control for throttle // TODO: 3 move all this PID crap into Actuators.cpp // TODO: 3 probably should do KF or something for speed/dist; need to address GPS lag, too // if nothing else, at least average the encoder speed over mult. samples if (desiredSpeed <= 0.1) { setThrottle( config.escZero ); } else { // PID loop for throttle control // http://www.codeproject.com/Articles/36459/PID-process-control-a-Cruise-Control-example float error = desiredSpeed - nowSpeed; // track error over time, scaled to the timer interval integral += (error * speedDt); // determine the amount of change from the last time checked float derivative = (error - lastError) / speedDt; // calculate how much to drive the output in order to get to the // desired setpoint. int output = config.escZero + (config.speedKp * error) + (config.speedKi * integral) + (config.speedKd * derivative); if (output > config.escMax) output = config.escMax; if (output < config.escMin) output = config.escMin; //fprintf(stdout, "s=%.1f d=%.1f o=%d\n", nowSpeed, desiredSpeed, output); setThrottle( output ); // remember the error for the next time around. lastError = error; } speedDt = 0; // reset dt to begin counting for next time control_count = CTRL_SKIP; } ////////////////////////////////////////////////////////////////////////////// // DATA FOR LOGGING ////////////////////////////////////////////////////////////////////////////// // Periodically, we enter a new SystemState into the FIFO buffer // The main loop handles logging and will catch up to us provided // we feed in new log entries slowly enough. if (--log_count == 0) { // Copy data into system state for logging inState++; // Get next state struct in the buffer inState &= SSBUF; // Wrap around ssBufOverrun = (inState == outState); // // Need to clear out encoder distance counters; these are incremented // each time this routine is called. state[inState].lrEncDistance = 0; state[inState].rrEncDistance = 0; // // need to initialize gps data to be safe // state[inState].gpsLatitude = 0; state[inState].gpsLongitude = 0; state[inState].gpsHDOP = 0; state[inState].gpsCourse = 0; state[inState].gpsSpeed = 0; state[inState].gpsSats = 0; log_count = LOG_SKIP; // reset counter } // Log Data Timestamp state[inState].millis = timestamp; // TODO: 3 recursive filtering on each of the state values state[inState].voltage = sensors.voltage; state[inState].current = sensors.current; for (int i=0; i < 3; i++) { state[inState].m[i] = sensors.m[i]; state[inState].g[i] = sensors.g[i]; state[inState].a[i] = sensors.a[i]; } state[inState].gTemp = sensors.gTemp; state[inState].gHeading = history[now].hdg; state[inState].lrEncSpeed = sensors.lrEncSpeed; state[inState].rrEncSpeed = sensors.rrEncSpeed; state[inState].lrEncDistance += sensors.lrEncDistance; state[inState].rrEncDistance += sensors.rrEncDistance; //state[inState].encHeading += (state[inState].lrEncDistance - state[inState].rrEncDistance) / TRACK; state[inState].estLatitude = here.latitude(); state[inState].estLongitude = here.longitude(); state[inState].estX = history[now].x; state[inState].estY = history[now].y; state[inState].bearing = bearing; state[inState].distance = distance; state[inState].nextWaypoint = nextWaypoint; state[inState].gbias = gyroBias; state[inState].errAngle = biasErrAngle; state[inState].leftRanger = sensors.leftRanger; state[inState].rightRanger = sensors.rightRanger; state[inState].centerRanger = sensors.centerRanger; state[inState].crossTrackErr = cte; // Copy AHRS data into logging data //state[inState].roll = ToDeg(ahrs.roll); //state[inState].pitch = ToDeg(ahrs.pitch); //state[inState].yaw = ToDeg(ahrs.yaw); // increment fifo pointers with wrap-around prev = now; inc(now); // timing tReal = timer.read_us() - tReal; ahrsStatus = 1; }