Turning an Automated System into an Autonomous system using Model-Based Design Autonomous Tech Conference 2018

Transcription

1 Turning an Automated System into an Autonomous system using Model-Based Design Autonomous Tech Conference 2018 Asaf Moses Systematics Ltd., Technical Product Manager 1

2 Autonomous System A system that functions independently or in a supervised manner and operates under conditions of uncertainty, in an unknown and unpredictable dynamic environment Platform Sense Mapping & Localization Planning Perception Planning Physical System + Environment Control Sensing & Perception Path Following (Control) Connect 2

3 Model-Based Design (MBD) Systematic use of models throughout the development process RESEARCH REQUIREMENTS DESIGN Environmental Models Scope: Electronic systems that interact with the physical world C, C++ Mechanical Control Algorithms Supervisory Logic IMPLEMENTATION CUDA VHDL Verilog Electrical Structured Text TEST & VERIFICATION MCU DSP GPU FPGA ASIC PLC, PAC INTEGRATION 3

4 Automated Driving - Sensors Definition Sense Camera Radar-based object detector Vision-based object detector Lidar Lane detector Inertial Measurement Unit 4

5 Multi-object tracker to develop sensor fusion algorithms Time Multi-Object Tracker State State Covariance Detections Track Manager Tracking Filter Tracks Track ID Age Is Confirmed Time Is Coasted Measurement Measurement Noise Assigns detections to tracks Creates new tracks Updates existing tracks Removes old tracks Predicts and updates state of track Supports linear, extended, and unscented Kalman filters 19:27 Introduction to Automated Driving System Toolbox 5

6 6 Fuse detections with multi-object tracker

7 What is Deep Learning? Perception Deep learning is a type of machine learning that performs end-to-end learning by learning tasks directly from images, text, and sound. 1. Train a Deep Neural Network from Scratch 2. Fine-tune a pre-trained model (transfer learning) 7

8 Network Analysis Network Analyzer Confusion Matrix 8

9 R-CNN Vs. Semantic Segmentation Regions with Convolutional Neural Network Features (R-CNN) Semantic Segmentation using SegNet 9

10 Path Planning Planning Map Initial Pose Final Pose Planner System Level Representation Path Controller Actuator Commands Probabilistic Roadmaps (PRM) Algorithm Application Area MATLAB Implementation PRM Ground Robots % Only for differential robots >> robotics.prm RRT/RRT* ADAS, Manipulators % Only for vehicles (3D) >> pathplannerrrt % RST % RST in future release %% Implicit Create A* search PRM path planner ADAS, Ground Robots planner Hybrid A* = robotics.prm; planner.map = og; Polynomial Interpolation Automotive, Ground Robots, planner.numnodes = 600; Manipulators planner.connectiondistance = 5; show(originalog); Model-Predictive Control hold on; Automotive, Humanoids % MPC Toolbox show(planner, 'Map', 'off'); >> mpc % General functions >> spline, pchip, Rapidly-exploring random tree (RRT) 10

11 Path Control Control Map Initial Pose Final Pose Planner System Level Representation Path Controller Actuator Commands x t 0 y(t 0 ) x t 1 y(t 1 ) x t N y(t N ) Path Controller x t y(t) Vehicle Model 12

12 Vehicle Modeling y w x t+1 = x t + v cmd cos θ t y t+1 = y t + v cmd sin θ t θ t+1 = θ t + w cmd t x y x 13

13 Path Control Explore Built-In Algorithms Angular velocity to go from current position to the lookahead Designing a system object so you can use your algorithm interchangeably in MATLAB and Simulink: - Cross tracking errors - Closest point to a line - Lookahead instance % Create a pure pursuit object pp = robotics.purepursuit; % Assign a sequence of waypoints pp.waypoints = [0 0;1 1;3 4]; % Compute control inputs for pose [v, w] = pp([0 0 0]); 14

14 Multi-domain Simulation Connect Robotics Algorithm Low-Level Control Robot Simulation 15

15 16 Physical Modeling Environment

16 17 Sign Recognition with Collision Avoidance

17 Vehicle Dynamics Automotive Aerospace Game engine 18

18 Extensive Hardware Support & Customization MATLAB & Simulink Simulink Coder Embedded Coder HDL Coder GPU Coder Mass market low cost hardware Rapid prototyping, student competitions, entry boards Industrial strength devices High performance SoCs Optimized CUDA code for deep learning 19

19 Autonomous Systems Technologies - Summary Position Estimation SLAM Map representation Occupancy grid State estimation Mission Planning Decision making Path and motion planning Collision avoidance Path following Platform Localization Planning Sense Perception Connectivity Planning Perception Controls Control Connect Scene Understanding Sensor fusion Deep learning Object detection Object tracking Sensors Cameras RADAR LIDAR GPS Commands Motion Control Actuators Controls Feedback controls Mode logic Gain scheduling Trajectory tracking 20

20 Contact Us: Systematics Ltd. Tel: Asaf Moses Seminars: Courses: Webinars: Thank You!