Mismatching dimension for field "x" with dimension (np.int32(3), np.int32(0)) (you have (2, 0))

MPC Discussions
,
1

Hi :wave:

Hi guys, I have a question regarding the calculation of next states in MPC simulation model. For instance, I have two states variables (y, yaw) defined by MX and 1 control (kappa) input in internal model and get the prediction states which contains these two(y, yaw). And then I passed them to the simulation model to get the update of them. But the simulation model I set contains 3 states (x, y, yaw) and 2 control inputs (delta, acc). When I’m gonna call the function below to calculate the next states.
The error turns out:

raise Exception(f'AcadosSimSolver.set(): mismatching dimension' \
Exception: AcadosSimSolver.set(): mismatching dimension for field "x" with dimension (np.int32(3), np.int32(0)) (you have (2, 0)).

def sim_x_next_compute(x, u, sim_x_next_integrator):
    # Set initial state
    sim_x_next_integrator.set("x", x)
    sim_x_next_integrator.set("u", u)
    # Solve the simulation
    status = sim_x_next_integrator.solve()
    if status != 0:
        raise Exception('acados returned status {}. Exiting.'.format(status))
    # Get the solution
    x_next_sim = sim_x_next_integrator.get('x')

    return x_next_sim

Should I always keep the same dimensions for both models? Only that could I be able to continue the simulation?

Best regards
zikhuan

2

Hi :wave:

in a standard setting, the simulator and the model used by the controller would be the same. Compare e.g. the minimal closed-loop example.

Why do you need different models for the simulator and the controller?

Best, Katrin

3

Hi Katrin,

thanks for your reply! I would like to try more complex simulation model since the prediction model only focuses on the y direction and yaw, which means it aims to minimize the lateral deviation. So you mean the two models are basically the same, especially the states?

Best

4

Have the two models the same control inputs?

5

Nope, dimension is also different. I would like to simulate in different ways. Like adding x direction states, and trying another different control inputs. Do you think it makes sense? Or at least the control inputs should be consistent.

6

It’s of course possible to have an integrator which uses a different model with potentially more states and controls. But then you need a way to come up with the extra controls which are not provided by the solver.

1 Like