Where Should We Place LiDARs on the Autonomous Vehicle? - An Optimal Design Approach

Abstract

Autonomous vehicle manufacturers recognize that LiDAR provides accurate 3D views and precise distance measures under highly uncertain driving conditions. Its practical implementation, however, remains costly. This paper investigates the optimal LiDAR configuration problem to achieve utility maximization. We use the perception area and nondetectable subspace to construct the design procedure as solving a min-max optimization problem and propose a bio-inspired measure -volume to surface area ratio (VSR) -as an easyto-evaluate cost function representing the notion of the size of the non-detectable subspaces of a given configuration. We then adopt a cuboid-based approach to show that the proposed VSRbased measure is a well-suited proxy for object detection rate. It is found that the Artificial Bee Colony evolutionary algorithm yields a tractable cost function computation. Our experiments highlight the effectiveness of our proposed VSR measure in terms of cost-effectiveness configuration as well as providing insightful analyses that can improve the design of AV systems.

Perception Entropy: A Metric for Multiple Sensors Configuration Evaluation and Design

“…Algorithm 1: Optimizing Installation Placement Given Sensor Selection Input: initial placement q 0 , initial neighborhood N , neighborhood lower bound N 0 , decay factor k Output: optimal placement q with entropy H min q = q 0 , H min = H(S|M, q 0 ) uniform sample perception space as set S while N > N 0 do random sample sensor placement around q within neighborhood N as set Q for q in Q do H total , p total = 0 for s in S do m = f (s, q) apply early fusion on m to obtain m f used and estimate AP calculate σ based on estimated AP apply late fusion on σ to obtain σ f used H(X|m, q) = 2ln(σ f used ) + 1 + ln(2π) H total = H total + p S (s)H(S|m, q) p total = p total + p S (s) end if H total /p total > H min then q = q, H min = H total /p total end end N = k • N end The running time of our algorithm is less than a second for a specific configuration and the whole optimization takes around an hour, which is much faster than previous methods [3], [4].…”

“…However, this method suffers from the curse of dimensionality and computation cost, which makes it unpractical for the real application. Liu et al [4] extended this work and proposed a new bionic metric named volume to surface area ratio (VSR) to analyze the tradeoffs between perception capability and design cost, but their modeling structure is still complex for real-world application and hardly reaches the global optimum despite simplifying the representation. Besides, there seems little relationship between their bionic metric VSR and the performance of perception algorithms, which also severely restricts the deployment of this approach to practical AVs' design.…”

“…To simplify the sensor configuration problem and simulation calculation, it is common to set up a perception space where the perception system is operated [3], [4]. We only consider the perception potential within this space.…”

“…Thus, the approaches are too rough to quantitatively evaluate the perception abilities of the sensor configuration. In addition, some works are focusing on the LiDAR configuration approaches [3], [4]. However, these kinds of methods cannot be applied to other kinds of sensors.…”

Optimal Placement of Roadside Infrastructure Sensors towards Safer Autonomous Vehicle Deployments

“…The sensing and perception capabilities of an AV by means of on-board sensors alone, may be limited and insufficient for urban environments, due to occlusions or limited range, even when their placement on the AV is optimal [2]. In contrast, combining the data from an AV's on-board sensors with that received from infrastructuremounted sensors using Vehicle-to-Infrastructure (V2I) communication is a way to enhance safety and performance.…”