Autonomous operation using deep spatio-temporal learning
A system and method to perform autonomous operation of a vehicle include obtaining one or more image frames for a time instance t from corresponding one or more sensors. Processing the one or more image frames includes performing convolutional processing to obtain a multi-dimensional matrix x t . The method includes operating on the multi-dimensional matrix x t to obtain output h t . The operating includes using an output h t−1 of the operating for a previous time instance t−1. The method also includes post-processing the output h t to obtain one or more control signals to affect operation of the vehicle.
1. A method of performing autonomous operation of a vehicle, the method comprising:
obtaining one or more image frames for a time instance t from corresponding one or more sensors;
processing the one or more image frames, the processing including performing convolutional processing to obtain a multi-dimensional matrix x t ;
operating on the multi-dimensional matrix x t to obtain output h t , wherein the operating includes using an output h t−1 of the operating for a previous time instance t−1; and
post-processing the output h t to obtain one or more control signals to affect operation of the vehicle.
2. The method according to claim 1 , wherein the processing the one or more image frames includes normalizing each of the one or more image frames prior to the performing the convolutional processing.
3. The method according to claim 1 , wherein the performing the convolutional processing includes performing N successive convolutions with N kernel matrices, wherein the kernel matrices have a random value and a pre-defined size and dimension initially.
4. The method according to claim 1 , wherein the operating on the multi-dimensional matrix x t using the output h t−t of the operating for the previous time instance t−1 includes performing computations of non-linear functions using weights.
5. The method according to claim 4 , further comprising comparing the one or more control signals respectively with one or more actual control signals during a training mode.
6. The method according to claim 5 , further comprising optimizing the weights based on the comparing.
7. The method according to claim 1 , wherein the post-processing the output h t to obtain the one or more control signals includes converting the output h t to a single dimensional vector and converting the single dimensional vector to the one or more control signals.
8. The method according to claim 1 , further comprising providing the one or more control signals to a vehicle control system of the vehicle.
9. The method according to claim 8 , further comprising providing the one or more control signals to an autonomous steering system of the vehicle.
10. The method according to claim 1 , wherein the one or more control signals includes one or more of a steering control signal, throttle level control signal, transmission control signal, or braking signal control signal.
11. A system to perform autonomous operation of a vehicle, the system comprising:
one or more sensors configured to obtain one or more image frames for a time instance t;
a controller configured to process the one or more image frames including performing convolutional processing to obtain a multi-dimensional matrix x t , to operate on the multi-dimensional matrix x t to obtain output h t , wherein operating includes using an output h t−1 of the operating for a previous time instance t−1, and to post-process the output h t to obtain one or more control signals to affect operation of the vehicle.
12. The system according to claim 11 , wherein the controller is further configured to process the one or more image frames by normalizing each of the one or more image frames prior to the performing the convolutional processing.
13. The system according to claim 11 , wherein the controller performing the convolutional processing includes performing N successive convolutions with N kernel matrices, wherein the kernel matrices have a random value and a fixed size and dimension initially.
14. The system according to claim 11 , wherein the controller is further configured to operate on the multi-dimensional matrix x t using the output h t−1 of the operating for the previous time instance t−1 by performing computations of non-linear functions using weights.
15. The system according to claim 14 , wherein the controller is further configured to compare the one or more control signals respectively with one or more actual control signals during a training mode.
16. The system according to claim 15 , wherein the controller is further configured to optimize the weights based on the comparing.
17. The system according to claim 11 , wherein the controller is further configured to post-process the output h t to obtain the one or more control signals by converting the output h t to a single dimensional vector and converting the single dimensional vector to the one or more control signals.
18. The system according to claim 11 , wherein the controller is further configured to provide the one or more control signals to a vehicle control system of the vehicle.
19. The system according to claim 18 , wherein the vehicle control system is an autonomous steering system of the vehicle.
20. The system according to claim 11 , wherein the one or more control signals is one or more of a steering control signal, throttle level control signal, braking control signal, or transmission control signal.