Self-driving scenario-based verification platform

ZIPC GARDEN

CATS

Challenge

  • Extensive running tests are required to secure the quality of a self-driving system.
  • It is impossible to guarantee all scenes are covered.

Countermeasures (Implemented Content)

  • For targeted use cases such as scenes of junctions and accidents, a comprehensive scenario-based approach implementing verification should be introduced.
  • To introduce virtual simulation environment with parallel batch-processing for many scenarios.

GARDEN verifies self-driving software based on scenarios.It is a comprehensive verification environment (platform) generating scenarios automatically and realizing precise and efficient verification with various simulators.

1. Scenario-based approach is introduced to make scenarios from accumulated knowledge.

By extracting critical scenarios from "Road Traffic Law", "Guidelines", "Driving Data", "Accident Data", etc., it generates scenarios automatically based on rules for road signs and dynamic parameters.

2. The auto-generated extensive scenarios are narrowed down into feasible numbers with software test methods and rule models.

The logical combination is extracted with N-wisemethod.

3. The scenarios are executed by linking with UE4, CarSIM, MATLAB/Simulink, etc., and tests are implemented in a high-quality virtual environment.

In addition to high-quality visual images, you can reduce verification man-hours in real space by generating point group data through LiDARfunction.


Challenge to embedded software development problem

ZIPC GARDEN Concept movie

Rule-based platform for edge intelligence

ZIPC R&B

CATS

What challenge does it solve?

For example, suppose you want to send different alerts to a driver from a hot country, not familiar with snowy roads, and a driver from a cold country, familiar with snowy roads. Table 1 and Table 2 are decision tables modeled with the editor of ZIPC R&B. Table 1 shows the road condition is categorized into 3 conditions of Dry, Wet and Snow and corresponding rules for each condition are called. Table 2 shows different rules for the alert when a rut is found on a snowy road depending on whether the driver has experience on a snowy road or not. On the simulation display of ZIPC R&B, you can check how alerts are shown on the Head Up Display (HUD)( Figure1).

Summary

We provide it as a platform with a rule-based engine with a small footprint, which can be embedded into an edge, and with MBD(Model-Based Development ) tool modeling CEP (Complex Event Processing) in state transition (Product Name: ZIPC R&B) . The target areas are all edges with intelligence, especially, it is attracting attention from businesses in the automotive area including ADAS (Advanced Driver-Assistance Systems) and already has a proven track record.

ZIPC R&Bのシミュレーション画面(Figure 1)ZIPC R&B Simulation Screen

路面状況ディシジョンテーブル(Table 1) Road Condition Decision Table

わだち警告ディシジョンテーブル(Table 2)Rut Alert Decision Table

CATS