An Optimal LiDAR Configuration Approach for Self-Driving Cars
This work addresses a domain-specific problem for self-driving car researchers by offering a guideline to maximize LiDAR utility, but it appears incremental as it builds on existing optimization techniques without claiming broad breakthroughs.
The paper tackles the problem of optimizing LiDAR configuration (placement and angles) for self-driving cars to improve object detection performance, and the proposed method is validated through simulations, though no concrete performance numbers are provided.
LiDARs plays an important role in self-driving cars and its configuration such as the location placement for each LiDAR can influence object detection performance. This paper aims to investigate an optimal configuration that maximizes the utility of on-hand LiDARs. First, a perception model of LiDAR is built based on its physical attributes. Then a generalized optimization model is developed to find the optimal configuration, including the pitch angle, roll angle, and position of LiDARs. In order to fix the optimization issue with off-the-shelf solvers, we proposed a lattice-based approach by segmenting the LiDAR's range of interest into finite subspaces, thus turning the optimal configuration into a nonlinear optimization problem. A cylinder-based method is also proposed to approximate the objective function, thereby making the nonlinear optimization problem solvable. A series of simulations are conducted to validate our proposed method. This proposed approach to optimal LiDAR configuration can provide a guideline to researchers to maximize the utility of LiDARs.