Demonstration of Safe Area Estimation by Trajectory Prediction Using Nonparametric Bayesian Inverse Reinforcement Learning
Automotive World 2020 (January 2020)
Estimation of safe area using trajectory prediction of other vehicles in an intersection scene.
Inverse reinforcement learning is used to predict the trajectories of other vehicles and humans.
By using a nonparametric bayesian-based algorithm as an elemental technology, the following advantages can be realized.
Predictions close to human senses are possible
Requires less training data than conventional DeepLearning or probabilistic models
Can be processed by edge devices
Possible to analyze what is used to output inference results
No need for humans to provide parameters; the model itself makes the decision
Edge device: NVIDIA Jetson AGX Xavier
Green: Trajectory prediction results for other vehicles