Starting PID
Traceback (most recent call last):
File "/home_net/malici/Downloads/lmpcugo/RacingLMPC/src/main.py", line 107, in
ClosedLoopDataPID = pickle.load(file_data)
UnicodeDecodeError: 'ascii' codec can't decode byte 0x9e in position 2: ordinal not in range(128)

Hi Ugo,

I faced with the following error:

  File "main.py", line 165, in <module>
    LMPCSimulator.Sim(ClosedLoopLMPC, LMPController, LMPCOpenLoopData)
  File "fnc/SysModel.py", line 37, in Sim
    Controller.solve(x[i, :])
  File "fnc/LMPC.py", line 96, in solve
    Qfun_SelectedTot    =  np.append(Qfun_SelectedTot, Qfun_Selected, axis=1)
  File "/usr/lib/python2.7/dist-packages/numpy/lib/function_base.py", line 4575, in append
    return concatenate((arr, values), axis=axis)
IndexError: axis 1 out of bounds [0, 1)

It seems that you should use axis=0 instead of axis=1 in line 96 of LMPC.py

Best.
Alex

(base) C:\dev\Projects_Work\RacingLMPC\src>python main.py
Starting PID
Traceback (most recent call last):
File "main.py", line 107, in
ClosedLoopDataPID = pickle.load(file_data)
ModuleNotFoundError: No module named 'Classes\r'

In PathfollowingLTIMPC.py -> should be as below:

# Build matrices for optimization (Convention from http://www.mpc.berkeley.edu/mpc-course-material, Optimal Control Handout, page 4-43, construction of a QP without sub)
# We are going to build our optimization vector z \in \mathbb{R}^((N+1) \dot n + N \dot d), note that this vector
# stacks the predicted trajectory x_{k|t} \forall k = t, \ldots, t+N+1 over the horizon and
# the predicted input u_{k|t} \forall k = t, \ldots, t+N over the horizon

Hi Rosolia, I am Hao, now studying autonomous racing vehicle in UCB. Your LMPC is amazing while I found a possible error in the fnc/PathFollowingLTVMPC.py, line 279, where you linearize the epsi/ey.
depsi_ey = - dt * (vx * np.cos(epsi) - vy * np.sin(epsi)) / (den**2) * cur * (-cur)
However, according to the kinetic model, which you write in the "Learning how to autonomously race a car: a predictive control approach"

I think this equation should be minus rather than positive based on fractional derivative. So I try to change this equation and finally found the simulator runs faster.(6.0s before, 5.8s after).
I am not quite sure if there is a minor error, but I hope this comment can help you. In all, thank you for sharing the LMPC. If you don't mind, I want to implement the LMPC in my project in UCB.

Starting PID
===== PID terminated
Starting MPC
===== MPC terminated
Starting TV-MPC
===== TV-MPC terminated
Starting LMPC
Traceback (most recent call last):
File "main.py", line 173, in
LMPCSimulator.Sim(ClosedLoopLMPC, LMPController, LMPCOpenLoopData)
File "fnc/SysModel.py", line 37, in Sim
Controller.solve(x[i, :])
File "fnc/LMPC.py", line 117, in solve
res_cons, feasible = osqp_solve_qp(sparse.csr_matrix(M), q, sparse.csr_matrix(F), b, sparse.csr_matrix(G), np.add(np.dot(E,x0),L[:,0]))
File "fnc/LMPC.py", line 426, in osqp_solve_qp
if res.info.status_val != osqp.constant('OSQP_SOLVED'):
AttributeError: 'OSQP' object has no attribute 'constant'

It's not clear how the control command are calculated https://github.com/urosolia/RacingLMPC/blob/master/src/fnc/Utilities.py#L67

Hi Ugo,

I was trying to implement your LMPC on the F1/10 autonomous car but had some trouble understanding the LMPC Relaxation part in your paper.

I'm a bit confused about why the time index is up to 4N. By doing this relaxation are you effectively trying to convert the problem into a QP?
I would highly appreciate if you could give any more insights regarding the idea behind the relaxation. Thank you!

Yuwei

Hi,
i tried the following new map:

    self.slack = 0.45
    self.halfWidth = 0.8
    self.lengthcurve = 4
    spec = np.array([[3/2, 0],
                     [self.lengthcurve/ 2 , self.lengthcurve /np.pi],
                     [self.lengthcurve / np.pi * (3.5), 0],
                     [self.lengthcurve /2,self.lengthcurve / np.pi],
                     [4/2, 0],
                     [self.lengthcurve  , self.lengthcurve / np.pi],
                     #[1/2 + 0.1, 0],
                     [self.lengthcurve, - self.lengthcurve / np.pi],
                     #[1/2, 0],
                     [self.lengthcurve / 2 *3, - 3* self.lengthcurve / np.pi],
                     [self.lengthcurve, self.lengthcurve / np.pi],
                     #[1-0.04647909, 0],
                     [self.lengthcurve/ 2 * 4.5, 4.5 * self.lengthcurve / np.pi]])

But then it fails in the first iteration of LMPC with the raise:
Starting PID
===== PID terminated
Starting MPC
===== MPC terminated
Starting TV-MPC
===== TV-MPC terminated
Starting LMPC
Traceback (most recent call last):
File "C:\Users\jonas\AppData\Local\Programs\Python\Python39\lib\site-packages\cvxopt\misc.py", line 1429, in factor
lapack.potrf(F['S'])
ArithmeticError: 1

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "C:\Users\jonas\AppData\Local\Programs\Python\Python39\lib\site-packages\cvxopt\coneprog.py", line 2010, in coneqp
try: f3 = kktsolver({'d': matrix(0.0, (0,1)), 'di':
File "C:\Users\jonas\AppData\Local\Programs\Python\Python39\lib\site-packages\cvxopt\coneprog.py", line 1981, in kktsolver
return factor(W, P)
File "C:\Users\jonas\AppData\Local\Programs\Python\Python39\lib\site-packages\cvxopt\misc.py", line 1444, in factor
lapack.potrf(F['S'])
ArithmeticError: 1

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "C:\Users\jonas\PycharmProjects\RacingLMPC-master(alt)\src\main.py", line 179, in
main()
File "C:\Users\jonas\PycharmProjects\RacingLMPC-master(alt)\src\main.py", line 116, in main
xTVMPC_cl, uTVMPC_cl, xTVMPC_cl_glob, _ = simulator.sim(xS, mpc)
File "C:\Users\jonas\PycharmProjects\RacingLMPC-master(alt)\src\fnc/simulator\SysModel.py", line 34, in sim
Controller.solve(x_cl[-1])
File "C:\Users\jonas\PycharmProjects\RacingLMPC-master(alt)\src\fnc/controller\PredictiveControllers.py", line 117, in solve
self.computeLTVdynamics()
File "C:\Users\jonas\PycharmProjects\RacingLMPC-master(alt)\src\fnc/controller\PredictiveControllers.py", line 144, in computeLTVdynamics
Ai, Bi, Ci = self.predictiveModel.regressionAndLinearization(self.xLin[i], self.uLin[i])
File "C:\Users\jonas\PycharmProjects\RacingLMPC-master(alt)\src\fnc/controller\PredictiveModel.py", line 70, in regressionAndLinearization
Ai[yIndex, self.stateFeatures], Bi[yIndex, self.inputFeaturesVx], Ci[yIndex] = self.LMPC_LocLinReg(Q_vx, b_vx, self.inputFeaturesVx)
File "C:\Users\jonas\PycharmProjects\RacingLMPC-master(alt)\src\fnc/controller\PredictiveModel.py", line 172, in LMPC_LocLinReg
res_cons = qp(Q, b) # This is ordered as [A B C]
File "C:\Users\jonas\AppData\Local\Programs\Python\Python39\lib\site-packages\cvxopt\coneprog.py", line 4485, in qp
return coneqp(P, q, G, h, None, A, b, initvals, kktsolver = kktsolver, options = options)
File "C:\Users\jonas\AppData\Local\Programs\Python\Python39\lib\site-packages\cvxopt\coneprog.py", line 2013, in coneqp
raise ValueError("Rank(A) < p or Rank([P; A; G]) < n")
ValueError: Rank(A) < p or Rank([P; A; G]) < n

Does someone have an idea why this map is not working correctly?

Hello，I am studying your code at [https://github.com/urosolia/RacingLMPC].

But I had some questions when I was running your program.

I use a judgment condition to judge whether the state variable exceeds the boundary constraint in 'SysModels.py' when a new state variable is generated :
x[i + 1, :], x_glob[i + 1, :] = _DynModel(x[i, :], x_glob[i, :], u[i, :], np, ClosedLoopData.dt, self.map.PointAndTangent) if x[i+1,5]>=self.map.halfWidth: print 'Lateral distance',x[i+1,5], 'Road boundary',self.map.halfWidth

But when I ran the program, I found that your program goes beyond the road boundary sometimes :

Lateral distance 0.4191460527916872 Road boundary 0.4 Lateral distance 0.4324614096754442 Road boundary 0.4 Lateral distance 0.4138255604138284 Road boundary 0.4

Therefore, I am more curious about whether there is an error margin when the road width constraint is exceeded.

Or is there any way to keep it strictly within the road width constraint?

In addition, I noticed that the position of the car is expressed in the form of mass points. Therefore, when the particle of the car is located at the edge of the road width constraint, what about the body part of the car that exceeds the road width constraint?

I will be very grateful if you can answer my questions.

Thanks!

Hi Ugo,

Thank for sharing this code. It really helps to understand the paper. I saw your video that demonstrates the use of this algorithm with a BARC car. I'm wondering wethere you're using the camera as input state to control the car or do you have the map of the track as with the simulator here. If the demo uses a map, how the car is get localized during the race ?