The REAKT project (https://reakt.sh/ & https://www.schiene-m-l.de/) aims at developing new mobility concepts to reactivate rural rail lines. This project will evolve around developing an autonomous rail vehicle, the REAKTOR,  to flexibly provide on-demand service on single track lines. A prototype will be built in 1:32 scale for 45 mm gauge, as well as a full scale version for the railway track Malente-Lütjenburg.

(left) a railbike, the foundation of the upcoming autonomous draisine & (right) a stripped down LGB engine for a 45mm model track.

Overview of topics (WIP)

The solutions in all topics should be scalable for both demonstrators!

1. An end-user app and management system for on-demand train service [already assigned (App part only)]
    This topic may be split up into two theses/project participants (app & management)

• A mobile app to call a train to a desired location and communicate desired destination
• Locations drawn from GNSS and appropriate abstraction for 1:32 scale for station-less entry or predefined locations
• Management of multiple on-demand trains on a single track line
• Schedules for cooperative passenger pick up
• Provisions for passenger transfer on open track ("Begegnungsverkehr", see concept art here)
• The inter train communication concept maybe based on a central or decentralized structure

2. An autonomous train controller with risk analysis using STPA [already assigned]

• Control of an autonomous passenger train model (conceptually working for both demonstrators)
• Capability for passenger transfer on open track (safe docking procedure; "Begegnungsverkehr", see concept art here)
• Risk analysis for the controller using STPA (http://psas.scripts.mit.edu/home/get_file.php?name=STPA_handbook.pdf)
• Safe behavior model generation using PASTA (https://marketplace.visualstudio.com/items?itemName=kieler.pasta)
• Assumes preprocessed sensor input and destination determination (see other topics)

3. AI-based obstacle detection for autonomous train control using image recognition [already assigned]
    This topic will be jointly advised with the AG Distributed Systems

• Sensor processing of a train-mounted camera to detect objects (potential obstacles)
• Tasks will involve:
  • Sensor mounting on the demonstrator
  • Collection of data (images, videos)
  • Labeling of data to enable training (esp. for small scale model)
  • Training of AI
  • Evaluation of quality
  • Live testing
• Step-wise evaluation of the influence of vehicle speed on the detection quality
• Evaluate applicability and influence of training data due to different environments for the demonstrators (i.e. indoors vs. outdoors)
• (Optional) Trajectory detection to categorize safety threads of moving obstacles
• (Optional) Evaluate performance on different hardware, e.g. Rasberry Pi vs. AI hardware
• (Optional) Test and evaluate on the edge deployment
• For interfacing with the controller, the sensor should provide an assessment how safe the area in front of the train is, such that the controller can adjust its speed.
• Potential hardware (subject to changes):
  • https://www.raspberrypi.com/documentation/accessories/camera.html

  • https://www.axis.com/de-de/products/axis-p1455-le

4. Classic and AI-based distance sensing for autonomous train control using different sensors
    This topic will be jointly advised with the AG Distributed Systems

• Explore and evaluate different sensors and processing techniques for distance measuring in rail vehicles
• Compare quality, ranges, and reliability w.r.t speed and size (demonstrator)
• Sensors and approaches:
  1. Ultrasonic sensor
  2. Single camera with AI image recognition
  3. (Multiple cameras)
  4. (LiDAR)
  5. (Sensorfusion)
• Potential hardware (subject to changes):
  • https://www.elektronik-kompendium.de/sites/praxis/bauteil_ultrasonic-hcsr04p.htm
  • https://www.pi-shop.ch/lidar-ld06-lidar-module-with-bracket-entwicklungskit-fuer-raspberry-pi-sbc

  • https://www.blickfeld.com/de/produkte/cube-1/

5. A digital twin for an autonomous on-demand train service [already assigned]
 Note: Tight interfacing with other topics

• A digital twin for an autonomous passenger train
• A virtual rebuild of the track using an 3D engine (i.e. Unreal Engine)
• Simulation capabilities to test and replay behavior (virtual environment)
• (Optional) Monitoring system for the state and location of the real vehicle
• (Optional) Monitoring and economic analysis of on-demand service operation (integration/interfacing of management system)
• (Optional) Reliability analysis/statistics to ensure transparency of autonomous operation

6. Remote control for rail vehicles [already assigned]
This topic will be jointly advised with the AG Distributed Systems
Note: Tight interfacing with other topics

• Remote control of speed and brakes (interfacing with controller)
• Live streaming of camera data and other sensors
• Evaluation and setup of a wireless communication network with a high reliability
• (Optional) Construction of a remote control panel
• (Optional) Augmented reality integration to simulate training scenarios

7. A standalone sensor box for monitoring rail vehicles [already assigned]
This prototype will be tested (only) using the full-scale demonstrator and is intended for monitoring non-autonomous vehicles (not the REAKTOR)

• Development of a sensor array to monitor rail vehicle operation
• It should serve as a plugin solution inside the train's cockpit for monitoring operation and as preparation for autonomous control
• Design for wireless communication of collected data
• Possible sensors:
  • GPS
  • Accelerometer
  • Camera
• Analysis of data for autonomous driving
• Potential integration into digital twin infrastructure

Goals

• TBA for each topic individually

Scope

Bachelor's or Master's Thesis, or Master's Project, with varying requirements to scientific scope.

Related Work/Literature

• https://reakt.sh/
• https://www.schiene-m-l.de/

Involved Languages/Technologies

• TBA for each topic individually

Supervised by

Alexander Schulz-Rosengarten  
als@informatik.uni-kiel.de