direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Inhalt des Dokuments

Autonomous Truck Platoon: Team 2

Group members: Teodor Janez Podobnik, Lubna Yaqoob, René Kriegel, Cheng Peng, Ayusch Khajuria, Yuke Qian, (Lu Kang, Thai Huy Do Nguyen)

Project Overview

Build an autonomous truck fleet that delivers cargo efficiently.

Goals

The main objective or goal of this project is to develop and build an intelligent autonomous truck fleet that delivers cargo efficiently. The main goals of this project are:

  1. Pick up cargo at predefined locations
    • Trucks can recognize stops at each location
  2. Deliver cargo quickly to its destination
    • Plan routes as short as possible
  3. Use available trucks and space per truck efficiently
    • Reasonable distribution of cargoes
  4. Platoon where possible, economical and safe
    • Keep the safety distance between each Truck in Platoon
    • get the Saved fuel consumption when Platoon
  5. Avoid collisions
    • Emergency stop when meet obstacles
  6. A booking Interface for User

Requirements

To achieve the goal efficiently the requirements are segregated to three different parts (consisting its related module) that are defined below.

Environment

Build a Highway system in Webots:

  • multiple entries and exits of pickup and drop-off locations
  • At least one depot for unused Trucks to park
  • different routes to the destination

Robot

Build a robot that interacts with the Webots environment:

  • Only use the following types of sensors: ultrasonic distance sensor, camera, light sensor
  • At least three robots in the environment

Control

Use the components Webots controller, external controller and backend:

  • Use the general architecture from the overview presentation: implement a TCP socket server in C for the Webots controller Determine the position of robots to achieve the mission
  • Make the operation safe (communication failure, sensor failure…)
  • Verify/validate the operation
  • Delivery cargoes efficiency (shortest path, fuel consumption...)

Approach

To handle this project efficiently we used.

Project Management

We worked in a Agile based model, where we plan and preview things on weekly basis. We utilised a various quality assurance strategies to ensure the quality of our project, for example an automated deployment system to make sure the integrated piece code works well.

Development

As mentioned above, we worked in component based teams (Webots controller, external controller, back-end and front-end). Communication between these components is done through TCP/IP communication protocols. Up to first milestone we come up with some basic work on each component. Until the end of second milestone we achieved our requirements goals for environment and external controller. Up to the third milestone we established communication between external controller and back-end. Our Webots component is developed in C language, whereas external controller consist of Ada language. For back-end development we used Python, where as front-end is in React JS. A complete insight of each component will be discussed in system architecture section.

System architecture

Our project consists of several components shown in a diagram below (just like the building pieces that comes together in a good solution), including webots-controller, external controller, back-end and front-end.

System_components

Due to the complexity of environment, autonomous trucks cannot be controlled by just one controller. Therefore, every truck has Webots controller and an external controller, the backend processes all truck information. We use C-Program for Webots controller, Ada for external controller and Python for backend.

Software design

For example, a user request cargo delivery from 'Mitte', a pickup location of cargo, to 'Grunewald's location to drop its freight of 500KG weight. How this procedure will work in all components will be explained in detail in this section with the help of image below.

System_design

As shown in the image we have three components interacting with each other. 'frontend', 'backend' and 'external controller'. Let see how it works all together. At first step, user send a request from the website to deliver 500 KG tomatoes, pickup from Mitte and drop it to Grunewald. This is a frontend request, and this frontend request information gets converted to JSON configuration, which then gets parsed to the backend. Backend is a service dictionary that consist of multiple services. The backend does the calculations and triggers the trucks. So the backend will be like to calculate which truck is utmost optimal to perform this task. Backend will do the calculations, for example, which truck is the closest to Mitte and Grunewald, or which has enough cargo free space and so forth. So the backend will then trigger the suitable/right external controller, and it will do the job. At the same time, backend will only trigger these commands only to the available trucks, so it won't send this if the truck is busy. This means that it will only distribute tasks between the available trucks, and if the truck is busy, it isn't considered in the cargo distribution. This is indicated by the blue arrow 'Multicast Task'. Then we have the green two way arrows that indicate the the request/feedback. So backend keep track of all the trucks, for example, truck 1 have 100 kilo of free cargo and it is in Mitte, or may be truck2 has 300 kilo of free cargo and its current position is in Charlottenburg. So truck1 needs to know where truck2 is. So it requests backend information where truck2 is. and backend will provide feedback to truck. This is why this is a two way arrow that gives feedback and tells that where the truck2 is.

Frontend design

This directory frontend is developed in React App. React App is an officially supported way to create single-page React applications. It offers a modern build setup with no configuration. Originally, it is developed for user to get trucks information. We are consuming getTruckInfo method of backend api. On the frontend user is requesting to get a truck in the form of object to the api. For sending request to api, we have used axios to make an HTTP post request to the API. And if the response is success than this response is passed to the calling component/function. There are some services frontend provides:

  • Communication between backend and frontend
  • Sending request parameters to get truck information

frontend1

Backend design

We use Backend as Service Manager which will provide multiple service ready to be served on a client request. In our case clients will be External controllers, Frontend user. Services Backend provides are:

  • Distribution of cargo between external controllers on a Frontend request using shortest path algorithm and custom cost function
  • Current status of each truck (position, free cargo space, busy)
  • Platoning Manager

Environment design

Overview

The environment will be the design as a highway system. It consists of multiple asphalt streets with road marking in different colors, 4 pickup and drop-off locations and a depot for unused trucks to park.

Points of Interest

Points of interest are specific locations in the environment. It can help us get the position of our Trucks and accurately find the destination to deliver cargoes. There are totally three types of Points of Interest in our environment:

  • Depot: Depot is the home of all Trucks. They can park here when they are not processing a job. Also when the Trucks finished their task, they will back to depot and waiting for new task.
  • Pick up/Drop off Location: There are four Locations in the environment with different line color.
  • Two Highways: There are two highways in the environment with different line color. It is used as the platoon sections and sometimes also the shortest way to get to the destination.

Map

Road marking

The white lines, which are positioned in the middle of the road, are used for trucks to navigate the path on the main road. The short red lines in each pick-up / drop-off location are used to stop, so that the trucks can stop a few second to pick up or drop off cargoes. Other lines with different colors are used to enter the different locations.

Roadmarking_location

Robot design

The robot (Truck) is a simple rectangular prism. It has six rectangular faces. As can be seen in the image below, we use different colors for the different robots so that they are easily identifiable. The following hardware is used:

  • Rectangular prism body
  • 4 wheels (two on each side)
  • 3 sensors.
    • one camera
    • two distance sensor (infra-red)

Robot

Lane detection

The camera placed in front is used to detect different color lines on the road.

Front distance

The distance sensors are places at the front left and right sides of the Trucks to detect the distance between each Truck in platoon as well as obstacles on the road.

Platooning

We create a multiple communication between each Truck in platoon. Each Truck has an array of Server connections and Clients. And we declare that in our world there are max 5 Trucks + only 5 Trucks can Platoon at a Time for safety reason. The Leader has is connected through Clients to every following car, while the other Trucks have just two Client-connections: one with the Leader and one with the Truck in front. This assure more safety in case of communications errors. In a degraded mode messages can be forwarded through the other Truck. Also when the leader Truck have a break to stop, due to the communication, the follower Trucks will also stop.

controll_peer

Algorithm

Follow Path

The Follow path package is the core component for driving. It based on the camera image. The key to processing the image is to let the truck know which kind of data is the color we want to follow. Then use follow color white as an example:

  • Detect BGR value: We use color_diff function to compare each pixel with BGR value of white, and 255 as a Threshold is used. If the result is smaller than 255, it will determine that the pixel is white, and if it is bigger than 255, it will determine that the pixel is not white. After this processing, we can obtain the BGR value array. 1 means white and 0 means not white.

image_processing_BGR

   function color_diff (a : color_BGR; b : color_BGR) return Integer is
      diff : Integer := 0;
      d    : Integer;
   begin
      for i in 0 .. 2 loop
         d := a (i) - b (i);
         if d > 0 then
            diff := diff + d;
         else
            diff := diff - d;
         end if;
      end loop;
      return diff;
   end color_diff; 
  • Get steering angle: Our Trucks always align themselves along the lines they follow. This is done by comparing the sum of the average values of all pixels and the index of the centerline.

image_processing_steer

  • Follow multiple colors We define 4 pick up/drop off locations and two highways with different colors. Then Truck should follow a color sequence in the environment. So we use an array of colors transmit to backend, backend calculate the route and send back the route as an array to external controller. Then the Trucks can follow multiple colors.

Postion determine

We create a node graph of our environment, with the Depot, Pickup/Drop-off locations and crossroads as nodes and the roads as edges with their street length as weight. Every section has his own unique color. By getting the current color line, the real simulation time from Webots and the velocity of the Truck, we can directly know the truck is located on which street and how many meters it travelled on this street.

node_graph

Shortest path

We using Dijkstra's algorithm with the help of Node_graph to find the shortest path between two selected Nodes in our map.

Distribution Algorithm

This algorithm is distributing cargoes and assign cargoes to available Trucks to finish the task.

def get_distribution(self, clients_free, cargo_request, optimum):  
   distrib = {}  
   for truck in optimum:  
       cargo_free = clients_free[truck]  
       if cargo_request > float(cargo_free):  
           distrib[truck] = int(cargo_free)  
           cargo_request -= int(cargo_free)  
       else:  
           distrib[truck] = int(cargo_request)  
           cargo_request = 0  
   print('distribution', distrib)   
   return distrib

Formal Verification

We formally verified some parts of external controller. The use of Ada programming language to implement the external controller provide access to use of SPARK for formal verification. In our code we used SPARK for the verification of follow_path and for position_algorithm. We used Pre- and Post conditions and -contracts to ensure program integrity. By developing appropriate contracts it could be verified that the examined functions and procedures work as desired. The data flow analysis and integrity checks where conducted using the gnatprove program and resulted in a total of 52 analyses for only follow_path shown in the image as an example:

spark

Summary

In this project, we have reached all of our mandatory goals:

  • A booking interface
  • Autonomous Truck:
    • Platooning with 5 Trucks in the map
    • Recognize pickup and destination locations
    • Calculate route
    • Safe operation: Avoid collisions
    • Delivery cargoes efficiently: Find shortest path
    • Delivery cargoes economy: Calculate the fuel cost during platooning However, we also have some goals could not be achieved:
  • Platooning only with Communication without control the distance between each Truck

Lesson learned

In this project we have learned about the importance of teamwork and communication, and everyone in our group did a good job and gave its best. Also the use of CI to test, build and document the project helped us avoid a lot of effort for integrating the individual components. During the development process, we learned Webots modeling, Webots node settings, Webots internal controller, and also the usage of various sensors in Webots. In the process of writing internal controllers, external controllers, front-end and back-end, our programming ability is improved.

Future work

In the future, we will continue to finish and optimize the part of Platooning, we decided to add a PID controller to keep the distance between each Truck in a safety value. Also in our current environment, we have only one lane road, in the future we can add more lanes in our map and more functions like overtake.

Zusatzinformationen / Extras

Direktzugang

Schnellnavigation zur Seite über Nummerneingabe