Methods and processors for controlling steering of self-driving car

Abstract

A method and processor for training a Machine Learning Algorithm (MLA) for predicting steering wheel positions for controlling steering of a Self-Driving Car (SDC). The method comprises: acquiring a plurality of training data sets a given one of which including (i) a training input and (ii) a respective label, the training input being indicative of (i) a training target state and (ii) a second target state, the respective label being indicative of an actual steering wheel position at a respective moment in time for arriving from the training target state to a sequential target state of a training trajectory; and inputting the given training data set into the MLA for determining a predicted steering wheel position for arriving from the training target state at the respective moment in time to the sequential target state while taking into account that the training trajectory is to end with the second target state.

1 . A computer-implementable method of training a Machine Learning Algorithm (MLA) for predicting steering wheel positions for controlling steering of a Self-Driving Car (SDC), the method comprising:
 acquiring a plurality of training data sets associated with a training trajectory of the SDC,
 the training trajectory having a plurality of target states of the SDC, the plurality of target states includes (i) a first target state of the SDC, (ii) a set of intermediary target states, and (iii) a second target state of the SDC,
 the target trajectory (i) beginning with the SDC being in the first target state, (ii) continuing with the SDC being sequentially in the set of intermediary target states, and (iii) ending with the SDC being in the second target state,
 the training trajectory being associated with a plurality of actual steering wheel positions at different moments in time, a given actual steering wheel position at a given moment in time for arriving from a given target state of the training trajectory to a sequential target state of the training trajectory, 
 
 a given one of the plurality of training data sets including (i) a training input and (ii) a respective label,
 the training input being indicative of (i) a training target state and (ii) the second target state, 
  the training target state being at least one of (i) the first target state and (ii) one of the plurality of intermediary target states, 
  the respective label being indicative of the actual steering wheel position at a respective moment in time for arriving from the training target state to the sequential target state of the training trajectory; 
 
 
   inputting the given training data set into the MLA for determining a predicted steering wheel position for arriving from the training target state at the respective moment in time to the sequential target state while taking into account that the training trajectory is to end with the second target state;   determining a difference between the predicted steering wheel position for the training target state and the actual steering wheel position for the training target state; and   training the MLA based on the difference to predict an in-use steering wheel position for the SDC for a current in-use state of the SDC while taking into account that an in-use trajectory of the SDC is to end at a second in-use target state.   
     
     
         2 . The method of  claim 1 , wherein:
 the inputting comprises iteratively inputting respective training data sets from the plurality of training data sets into the MLA for determining respective predicted steering wheel positions; and wherein   the training comprises iteratively training the MLA based on differences between the respective predicted steering wheel positions and respective actual steering wheel positions.   
     
     
         3 . The method of  claim 1 , wherein the plurality of actual steering wheel positions associated with the training trajectory are in a form of a steering profile for the training trajectory, the steering profile having been built as a polynomial curve representative of a given steering profile of the SDC that allows the SDC to transition from the first target state to the second target state. 
     
     
         4 . A system for training a Machine Learning Algorithm (MLA) for predicting steering wheel positions for controlling steering of a Self-Driving Car (SDC), the system comprising at least one processor and at least one non-transitory computer-readable memory comprising instructions that, when executed by the at least one processor, cause the system to:
 acquire a plurality of training data sets associated with a training trajectory of the SDC,
 the training trajectory having a plurality of target states of the SDC, the plurality of target states includes (i) a first target state of the SDC, (ii) a set of intermediary target states, and (iii) a second target state of the SDC,
 the target trajectory (i) beginning with the SDC being in the first target state, (ii) continuing with the SDC being sequentially in the set of intermediary target states, and (iii) ending with the SDC being in the second target state,
 the training trajectory being associated with a plurality of actual steering wheel positions at different moments in time, a given actual steering wheel position at a given moment in time for arriving from a given target state of the training trajectory to a sequential target state of the training trajectory, 
 
 a given one of the plurality of training data sets including (i) a training input and (ii) a respective label,
 the training input being indicative of (i) a training target state and (ii) the second target state, 
  the training target state being at least one of (i) the first target state and (ii) one of the plurality of intermediary target states, 
  the respective label being indicative of the actual steering wheel position at a respective moment in time for arriving from the training target state to the sequential target state of the training trajectory; 
 
 
   input the given training data set into the MLA for determining a predicted steering wheel position for arriving from the training target state at the respective moment in time to the sequential target state while taking into account that the training trajectory is to end with the second target state;   determine a difference between the predicted steering wheel position for the training target state and the actual steering wheel position for the training target state; and   train the MLA based on the difference to predict an in-use steering wheel position for the SDC for a current in-use state of the SDC while taking into account that an in-use trajectory of the SDC is to end at a second in-use target state.