the problem of the thruster

[fbgdhn]

class thruster:
......
def compute(self, control_action: MultiRotorActions) -> MultiRotorActions:
"""Advance the thruster state one step.

    Applies saturation, chooses rise/fall tau per motor, computes mixing factor,
    and integrates with the selected kernel.

    Args:
        control_action: (num_envs, num_thrusters) commanded per-thruster thrust [N].

    Returns:
        (num_envs, num_thrusters) updated thrust state [N].

    """
    des_thrust = control_action.thrusts
    des_thrust = torch.clamp(des_thrust, *self.thrust_r)

    thrust_decrease_mask = torch.sign(self.curr_thrust) * torch.sign(des_thrust - self.curr_thrust)
    motor_tau = torch.where(thrust_decrease_mask < 0, self.tau_dec_s, self.tau_inc_s)
    mixing = self.mixing_factor_function(motor_tau)

    self.curr_thrust[:] = self._step_thrust(des_thrust, self.curr_thrust, mixing)

    self.computed_thrust = self.curr_thrust
    self.applied_thrust = torch.clamp(self.computed_thrust, self.min_thrust, self.max_thrust)

    control_action.thrusts = self.applied_thrust

    return control_action

The def thruster is used ' thrust_decrease_mask = torch.sign(self.curr_thrust) * torch.sign(des_thrust - self.curr_thrust)' to determine whether the drone is ascending or descending.This assumes that the value of 'self.curr_thrust' is positive.But when we try to descending the drone at a large acceleration,the value of 'self.curr_thrust' is negative.In this situation,the judgement is wrong.

Is my understanding correct?I hope you’ll share your thoughts.