Path prediction for a vehicle
A method and a system for predicting a near future path and an associated output control signal for a vehicle. Prediction sensor data, vehicle driving data, and road data are collected. An input control signal indicative of an intended driving action is received. The sensor data and the vehicle driving data are pre-processed to provide a set of object data comprising a time series of previous positions of a respective object relative the vehicle, a time series of the previous headings of the object, and a time series of previous velocities of the object. The object data, the road data, the vehicle driving data, the control signal, and the sensor data are processed in a deep neural network. Based on the processing in the deep neural network, a predicted path output and an output control signal are provided.
1. A method for predicting a near future path for a vehicle to thereby provide an associated output control signal, the method comprising:
collecting vehicle driving data indicative of at least a velocity and a yaw rate of the vehicle,
collecting sensor data indicative of the presence of objects in a vicinity of the vehicle,
collecting road data indicative of a roadway on a presently occupied road for the vehicle,
receiving an input control signal indicative of an intended driving action for the vehicle,
pre-processing the sensor data and the vehicle driving data to provide a set of object data comprising a time series of previous positions of a respective object relative to the vehicle, a time series of the previous headings of the object, and a time series of previous velocities of the object relative the vehicle,
processing the object data, the road data, the vehicle driving data, the control signal, and the sensor data in a deep neural network to predict the near future path for the vehicle,
based on the processing in the deep neural network, providing a predicted path output and an output control signal for executing a driving action according to the predicted path.
2. The method according to claim 1 , further comprising:
collecting map data indicative of a road environment and legal road characteristics,
processing the map data in the deep neural network to predict the near future path.
3. The method according to claim 1 , further comprising:
performing a safety check on the output control signal indicative of a driving action,
when the driving action is approved, processing the output control signal together with updated object data and updated road data in the deep neural network to predict a further near future path,
based on the processing in the deep neural network, providing an updated predicted path output.
4. The method according to claim 1 , further comprising:
performing a safety check on the predicted near future path,
based on the outcome of the safety check, executing the predicted near future path.
5. The method according to claim 1 , wherein in the deep neural network, the sensor data is passed through convolutional layers to provide low dimensional sensor data representations,
the object data and the road data is passed through fully connected layers to provide low dimensional object data representations and low dimensional road data representations.
6. The method according to claim 1 , wherein in the deep neural network, the sensor data, the object data, and the road data are passed through convolutional layers to provide respective low dimensional data representations.
7. The method according to claim 5 , wherein the low dimensional data representations are processed through additional layers to predict the output near future path and the output control signals.
8. The method according to claim 1 , wherein the deep neural network is trained by supervised learning based on target values and paths recorded from human drivers in traffic.
9. The method according to claim 1 , wherein the deep neural network is trained by un-supervised learning based on a definition of safe and natural driving.
10. The method according to claim 1 , wherein the time series includes position data and velocity data from the previous 0 to 1 second, 0 to 2 seconds, 0 to 3 seconds, 0 to 4 seconds, or 0 to 5 seconds.
11. A system for predicting a near future path for a vehicle to thereby provide an associated output control signal, the system comprising:
an out-looking sensor unit for obtaining sensor data indicative of the presence of objects in a vicinity of the vehicle and for capturing road data indicative of a roadway on a presently occupied road for the vehicle,
a vehicle sensor unit for collecting vehicle driving data indicative of at least a velocity and a yaw rate of the vehicle,
a safety control module for providing a control signal indicative of an intended driving action for the vehicle, and
processing circuitry configured to:
pre-process the sensor data and the vehicle driving data to provide a set of object data comprising a time series of previous positions of a respective object relative the vehicle, a time series of previous headings of the object, and a time series of previous velocities of the object relative the vehicle,
process the object data, the road data, the vehicle driving data, the control signal, and the sensor data in a deep neural network to predict the near future path for the vehicle,
based on the processing in the deep neural network, provide a predicted path output and an output control signal for executing a driving action according to the predicted path.
12. The system according to claim 11 , wherein the out-looking sensor unit comprises an image capturing device, a radar, a lidar, or a combination thereof.
13. The system according to claim 11 , further comprising:
a map unit configured to provide map data indicative of a road environment and legal present road characteristics, wherein the map data is used by the processing circuitry in the deep neural network to predict the near future path for the vehicle.
14. The system according to claim 11 , wherein the safety control module is configured to receive the output control signal and perform safety check of the indicated driving action,
wherein if the intended driving action is approved by the safety control module, the safety control module is configured to provide the output control signal to the processing circuitry to provide an updated predicted path output.