Page 12 - 卫星导航2021年第1-2合期

El‑Sheimy and Li Satell Navig (2021) 2:7 Page 2 of 23

by reducing congestion, accidents, energy consumption, Compared with industry and construction, the PLAN
and time consumption (Schönenberger 2019). Te huge accuracy requirements for autonomous driving are lower.
social and economic benefts promote the demand for However, the application scene is much larger and has
PLAN technology facing the autonomous driving and more complex changes; also, the cost is more restrictive.
mass consumer markets. Such factors increase the challenge of PLAN in autono-
mous driving. Te Society of Automotive Engineers
Classifcation of indoor navigation from market divides autonomous driving into L0 (no automation), L1
perspective (driver assistance), L2 (partial automation), L3 (condi-
PLAN technology is highly related to market demand. tional automation, which requires drivers to be ready to
Table 1 shows the accuracy requirements and costs of take over when the vehicle has an emergency alert), L4
several typical indoor PLAN applications. (high automation, which does not require any user inter-
In general, for the applications that require higher vention but is only limited to specifc operational design
accuracy, the facilities and equipment costs are corre- domains, such as areas with specifc facilities and High-
spondingly higher. In many scenarios (e.g., the mass-mar- Defnition maps (HD maps), and L5 (fully automation)
ket ones), the minimum equipment installation cost and (SAE-International 2016). In most situations, autono-
equipment cost are important factors that limit the scal- mous cars mean L3 and above. Tere is still a certain
ability of PLAN technology. distance from L5 commercial use (Wolcott and Eustice
Industry and construction require the PLAN accuracy 2014). An important bottleneck is that PLAN technol-
at the centimeter- or even millimeter-level. For example, ogy is difcult to meet the requirements in the entire
the accuracy requirements for machine guidance and environment.
deformation analysis are 1–5 cm and 1–5 mm, respec- Tere are various derivations and defnitions of the
tively. Te corresponding cost is in the $ 10,000 level accuracy requirement of autonomous driving. Table 2
(Schneider 2010). lists several of those derivations and defnitions.

Table 1 Accuracy requirements and costs of typical indoor PLAN applications
Application Accuracy requirement Cost
Industry and construction (Schneider 2010) Centimeter‑level to millimeter‑level $ 10,000 level
Autonomous vehicles (Basnayake et al. 2010; Decimeter‑level to centimeter‑level $ 1,000 level to $ 10,000 level
Levinson and Thrun 2010; NHTSA 2017; Reid
et al. 2019; Agency 2019; Stephenson 2016;
Nvidia 2020)
Indoor mapping (Cadena et al. 2016) Decimeter‑level to centimeter‑level $ 1,000 level
First responder (Rantakokko et al. 2010) Decimeter‑level in horizontal, foor‑level in height $ 1,000 level
Pedestrian applications (Dodge 2013) Meter‑level in horizontal, foor‑level in height Use existing consumer devices; infrastruc‑
ture deployment cost of $ 10 level per 100
2
m ‑level area
Cellular emergency (FCC 2015) 80% within 50 m Use existing cellular systems

Table 2 Derivations and defnitions of accuracy requirement for autonomous driving
Reference Analysis of accuracy requirement of autonomous cars
Research (Basnayake et al. 2010) Within 5 m, within 1.5 m, and within 1.0 m for which‑road, which‑lane, and where‑in‑lane, respectively, in V2X
applications
Report (NHTSA 2017) 1.5 m (1 sigma, 68%) tentatively for lane‑level information for safety applications
Research (Reid et al. 2019) For passenger vehicles operating, the bounds of lateral and longitudinal position errors are respectively 0.57 m
(95% probability in 0.20 m) and 1.40 m (95% probability in 0.48 m) on freeway roads, and both 0.29 m (95%
probability in 0.10 m) on local streets
Research (Levinson and Thrun 2010) Centimeter positioning accuracy with Mean Square Error (MSE) within 10 cm is sufciently accurate for public
roads
Report (Agency 2019) The accuracy of autonomous driving to be within 20 cm in horizontal and within 2 m in height
Research (Stephenson 2016) Active vehicle control in Advanced Driver Assistance Systems (ADAS) and autonomous driving applications
require an accuracy better than 0.1 m
Industry (Nvidia 2020) The goal is centimeter‑level