Perception Imitation: Towards Synthesis-free Simulator for Autonomous Vehicles
This work addresses the need for more economic and efficient simulators for autonomous driving by focusing on perception imitation, but it appears incremental as it builds on existing simulation and perception methods.
The paper tackles the problem of simulating perception results for autonomous vehicles without synthesizing sensor data, proposing a perception imitation method that directly models the behavior of learning-based perception models, and experiments show it is effective and can be applied smoothly in a new simulation route.
We propose a perception imitation method to simulate results of a certain perception model, and discuss a new heuristic route of autonomous driving simulator without data synthesis. The motivation is that original sensor data is not always necessary for tasks such as planning and control when semantic perception results are ready, so that simulating perception directly is more economic and efficient. In this work, a series of evaluation methods such as matching metric and performance of downstream task are exploited to examine the simulation quality. Experiments show that our method is effective to model the behavior of learning-based perception model, and can be further applied in the proposed simulation route smoothly.