Joint signature-virtual field representation

1 . A method of joint signature-virtual field representation of road elements for driving, the method comprising:

receiving, at a machine learning process associated with a vehicle, multiple sensed information units representing a road element captured at different points of time; generating, by the machine learning process, a joint representation of the road element that is a single unified representation that simultaneously encodes both a class indication of the road element and of a virtual field indication of an impact of the road element on the vehicle; wherein the virtual field indication represents a virtual force virtually applied by the road element on the vehicle based on a virtual physical model; wherein the class indication is distinguishable from the virtual field indication;

determining in real time in a driving of the vehicle, based on the joint representation, a driving related output affecting the vehicle, wherein the driving related output is a driving related decision with respect to the vehicle; wherein the determining of the driving related output comprises: (a) determining the virtual force virtually applied by the road element on the vehicle, (b) determining, based the virtual force, a virtual acceleration applied on the vehicle by the road element, (d) determining a desired acceleration of the vehicle based, at least in part, on the virtual acceleration applied on the vehicle by the road element, wherein the driving related output is used to have the vehicle autonomously drive at the desired acceleration; and

autonomously driving the vehicle by executing the driving related decision.

2 . The method according to claim 1 , wherein the machine learning process being trained using a classification loss and a virtual field loss.

3 . The method according to claim 1 , comprising training the machine learning process to generate the joint representation.

4 . The method according to claim 1 , comprising ignoring the joint representation of the road element when the class indication indicates that the road element is of a given class that has an impact on the vehicle that does not match the impact on the vehicle that is indicated by the virtual field indication.

5 . The method according to claim 1 , comprising storing the virtual field indication and the class indication in a single memory entry or in consecutive entries of a single data structure.

6 . The method according to claim 5 , comprising determining the desired acceleration based on the equivalent of Newton's second law.

7 . The method according to claim 1 , wherein the autonomously driving the vehicle is executed under a control of an autonomous driving control unit that comprises one or more processing circuits.

8 . A non-transitory computer readable medium for joint signature-virtual field representation of road elements for driving, the non-transitory computer readable medium stores commands executable by a processing circuit for: receiving, at a machine learning process associated with a vehicle, multiple sensed information units representing a road element captured at different points of time; generating, by the machine learning process, a joint representation of the road element that is a single unified representation that simultaneously encodes both a class indication of the road element and of a virtual field indication of an impact of the road element on the vehicle, wherein the virtual field indication represents a virtual force virtually applied by the road element on the vehicle based on a virtual physical model; wherein the class indication is distinguishable from the virtual field indication;

determining in real time in a driving of the vehicle, based on the joint representation, a driving related output affecting the vehicle, wherein the driving related output is a driving related decision with respect to the vehicle; wherein the determining of the driving related output comprises: (a) determining the virtual force virtually applied by the road element on the vehicle, (b) determining, based the virtual force, a virtual acceleration applied on the vehicle by the road element, (d) determining a desired acceleration of the vehicle based, at least in part, on the virtual acceleration applied on the vehicle by the road element, wherein the driving related output is used to have the vehicle autonomously drive at the desired acceleration; and

autonomously driving the vehicle by executing the driving related decision.

9 . The non-transitory computer readable medium according to claim 8 , wherein the machine learning process being trained using a classification loss and a virtual field loss.

10 . The non-transitory computer readable medium according to claim 8 , that further stores instructions executable by a processing circuit for training the machine learning process to generate the joint representation.

11 . The non-transitory computer readable medium according to claim 8 , that further stores instructions executable by the processing circuit for ignoring the joint representation of the road element when the class indication indicates that the road element is of a given class that has an impact on the vehicle that does not match the impact on the vehicle that is indicated by the virtual field indication.

12 . The non-transitory computer readable medium according to claim 8 , comprising storing the virtual field indication and the class indication in a single memory entry or in consecutive entries of a single data structure.

13 . The non-transitory computer readable medium according to claim 12 , wherein the determining of the desired acceleration is based on the equivalent of Newton's second law.

14 . The non-transitory computer readable medium according to claim 8 , wherein the autonomously driving the vehicle is executed under a control of an autonomous driving control unit that comprises one or more processing circuits.