Johann Christensen (né Lange)

Johann Christensen (né Lange)

he/him

Research associate at the Institute for AI Safety and Security at the German Aerospace Center (DLR)

Publications

You can also find my articles on my ORCiD or Google Scholar profile.

Journal Articles

Automated Scenario Generation from Operational Design Domain Model for Testing AI-Based Systems in Aviation

Published in CEAS Aeronautical Journal, 2024

Artificial Intelligence (AI) applications, including those in aviation, face regulatory challenges for safety assurance and certification. The EASA emphasizes the importance of aligning the Concept of Operations (ConOps) with the Operational Design Domain (ODD) for effective safety analysis. This study introduces a framework for generating and testing synthetic data to support ConOps. This scenario-based testing approach, combined with Model-Based Systems Engineering (MBSE), facilitates efficient verification and automated testing based on ODD definitions.

Recommended citation: Stefani, T., Christensen, J. M., Anilkumar Girija, A., Gupta, S., Durak, U., Köster, F., Krüger, T. and Hallerbach, S. "Automated Scenario Generation from Operational Design Domain Model for Testing AI-Based Systems in Aviation", in CEAS Aeronautical Journal, Sep. 2024.
Download Paper

Design and Adaptive Depth Control of a Micro Diving Agent

Published in IEEE Robotics and Automation Letters (RA-L), 2017

This letter describes the depth control of an autonomous micro diving agent (ADA), built from off-the-shelf components with open-source hardware and firmware. ADA serves as a testbed for depth controllers and a mobile sensor platform. A feedback linearization control law, enhanced with an adaptive fuzzy algorithm, addresses modeling inaccuracies and is suitable for ADA’s embedded hardware. Experiments in a wave tank show that the adaptive fuzzy controller effectively manages depth regulation and profile tracking, enabling ADA to maintain stability and exhibit orbital motions akin to water particles under wave influence.

Recommended citation: Bessa, W. M., Kreuzer, E., Lange, J., Pick, M. and Solowjow E. "Design and Adaptive Depth Control of a Micro Diving Agent", in IEEE Robotics and Automation Letters, vol. 2, no. 4, pp. 1871-1877, Oct. 2017, doi: 10.1109/LRA.2017.2714142.
Download Paper

Conference Papers

Advancing the AI-Based Realization of ACAS X Towards Real-World Application

Published in 36th IEEE International Conference on Tools with Artificial Intelligence (ICTAI), 2024

AI is increasingly applied in safety-critical domains like automotive and robotics, but its deployment in aviation remains limited due to stringent safety requirements. This work explores AI’s potential in aviation through the Airborne Collision Avoidance Systems X (ACAS X), including ACAS XA for vertical advisories and ACAS XU for horizontal advisories. Implementations of both variants for FlightGear demonstrate their effectiveness in avoiding near mid-air collisions (NMACs). An Operational Design Domain is defined for safety considerations, and simulation tests validate the successful use of advisory predictions for collision avoidance.

Recommended citation: Christensen, J. M., Anilkumar Girija, A., Stefani, T., Durak, U., Hoemann, E., Köster, F., Krüger, T. and Hallerbach, S. "Advancing the AI-Based Realization of ACAS X Towards Real-World Application", in 36th IEEE International Conference on Tools with Artificial Intelligence (ICTAI), Oct. 2024.
Download Paper | Download Slides

Towards the Monitoring of Operational Design Domains using Temporal Scene Analysis in the realm of Artificial Intelligence in Aviation

Published in 2024 IEEE/AIAA 43st Digital Avionics Systems Conference (DASC), 2024

AI in aviation, especially in safety-critical applications, demands strict safety measures. Defining and monitoring the Operational Design Domain (ODD) is crucial for ensuring AI system safety. This involves training the AI on substantial data and understanding its operating environment. This paper introduces a novel predictive ODD monitoring methodology using temporal scene analysis. By dividing scenarios into discrete scenes and creating synthetic scenarios, this approach enhances data for assessing AI performance and safety.

Recommended citation: Anilkumar Girija, A., Christensen, J. M., Stefani, T., Hoemann, E., Durak, U., Köster, F., Hallerbach, S. and Krüger, T. "Towards the Monitoring of Operational Design Domains using Temporal Scene Analysis in the realm of Artificial Intelligence in Aviation", in 2024 IEEE/AIAA 43st Digital Avionics Systems Conference (DASC), Sep. 2024.
Download Paper

Towards an Operational Design Domain for Safe Human-AI Teaming in the Field of AI-Based Air Traffic Controller Operations

Published in 2024 IEEE/AIAA 43st Digital Avionics Systems Conference (DASC), 2024

Advances in AI offer promising applications in aviation but pose safety challenges, particularly in Human-AI Teaming. The European Union Aviation Safety Agency (EASA) emphasizes safety in dynamic human-AI collaboration with shared goals and partial authority. This work adapts the Operational Design Domains (ODDs) concept to Air Traffic Control (ATC), defining an initial ODD for an AI-based team partner aiding conflict detection and resolution. Simulations test scenarios to evaluate ODD feasibility in ATC safety.

Recommended citation: Stefani, T., Jameel, M., Hunger, R., Gerdes, I., Anilkumar Girija, A., Christensen, J. M., Bruder, C., Köster, F., Hallerbach, S. and Krüger, T. "Towards Certifiable AI in Aviation: A Framework for Neural Network Assurance Using Advanced Visualization and Safety Nets", in 2024 IEEE/AIAA 43st Digital Avionics Systems Conference (DASC), Sep. 2024.
Download Paper

Towards Certifiable AI in Aviation: A Framework for Neural Network Assurance Using Advanced Visualization and Safety Nets

Published in 2024 IEEE/AIAA 43st Digital Avionics Systems Conference (DASC), 2024

AI’s growing role in technology necessitates stringent safety standards, especially in aviation where certification is regulated by strict laws. Ensuring AI system safety requires thorough understanding and proper tools. This work presents a Safety Net, utilizing sparse lookup tables (LUTs) to address neural network failures. By combining LUTs with neural networks, a certifiable system can be created, offering a framework for 100% reliability and effective real-time corrections during operation.

Recommended citation: Christensen, J. M., Zaeske, W., Anilkumar Girija, A., Friedrich, S., Stefani, T., Durak, U., Köster, F., Krüger, T. and Hallerbach, S. "Towards Certifiable AI in Aviation: A Framework for Neural Network Assurance Using Advanced Visualization and Safety Nets", in 2024 IEEE/AIAA 43st Digital Avionics Systems Conference (DASC), Sep. 2024.
Download Paper | Download Slides

Applying Model-Based System Engineering and DevOps on the Implementation of an AI-based Collision Avoidance System

Published in 34th Congress of the International Council of the Aeronautical Sciences (ICAS), 2024

The integration of AI in aviation offers advancements but requires stringent regulation to ensure safety. EASA introduced a W-model framework for AI, complementing the V-model, though it diverges from the DevOps cycle commonly used in AI development. To explore harmonization, an AI-based version of ACAS XA and ACAS XU has been used as a use case. Model-Based System Engineering (MBSE) facilitated managing complexity and improved stakeholder communication.

Recommended citation: Stefani, T., Christensen, J. M., Hoemann, E., Anilkumar Girija, A., Köster, F., Krüger, T. and Hallerbach, S. "Applying Model-Based System Engineering and DevOps on the Implementation of an AI-based Collision Avoidance System", in 34th Congress of the International Council of the Aeronautical Sciences, Sep. 2024.

Academic Theses

Obstacle avoidance-driven controller for safety-critical aerial robots

Published in Hamburg University of Technology (TUHH), 2019

This thesis introduces the Model-Predictive-Control-Barrier-Function (MPCBF), combining Control-Barrier-Functions (CBF) with Model-Predictive-Control (MPC). The MPCBF demonstrates superior performance over CBF, attributed to MPC’s extended time horizon. Applied to a quadrotor, a system requiring rapid, predictive control, MPCBF enabled effective obstacle avoidance, outperforming CBF due to handling relative obstacle speeds. The proposed approach is validated through experimental results.

Recommended citation: Lange, J. "Obstacle avoidance-driven controller for safety-critical aerial robots", in Hamburg University of Technology (TUHH), Sep. 2019, eprint: 2011.08178.
Download Paper

Preprints

Comparing Popular Simulation Environments in the Scope of Robotics and Reinforcement Learning

Published in arXiv.org, 2021

This letter benchmarks the performance of four popular simulation environments for robotics and reinforcement learning (RL), considering current industrial applications. The comparison evaluates various hardware configurations, from entry-level notebooks to high-performance dual-CPU servers, to optimize RL training time. Results show that single-core performance is critical, but multi-core systems allow for parallel simulations, enhancing overall performance.

Recommended citation: Körber, M., Lange, J., Rediske, S., Steinmann, S. and Glück, R. "Comparing Popular Simulation Environments in the Scope of Robotics and Reinforcement Learning", in arXiv preprint, March 2021, eprint: 2103.04616.
Download Paper

Supervised Theses

Ensuring the Trustworthy Development of AI Based Applications Compatible to Future EASA Regulations

Published in Ulm University of Applied Sciences (THU), 2025

This thesis advances EASA-compliant AI certification by introducing a methodology to define ConOps, OD, and ODD for AI-based aviation systems. It provides a structured approach to specifying these concepts, ensuring alignment with EASA’s learning assurance framework and enhancing trustworthiness, safety, and regulatory compliance in AI-driven aviation applications.

Recommended citation: Werner, F. "Ensuring the Trustworthy Development of AI Based Applications Compatible to Future EASA Regulations", Master's Thesis, Ulm University of Applied Sciences, January 2025.
Download Paper

Reward Shaping for Reinforcement Learning in the Aviation Industry

Published in Hamburg University of Technology (TUHH), 2022

This work advances robotic automation by applying deep reinforcement learning to a robotic gripper for door opening. It compares SAC and PPO algorithms, explores reward function design for improved learning behavior, and validates results in simulated environments. Findings highlight PPO’s efficiency and the critical role of reward shaping in training success.

Recommended citation: Wei, J. "Reward Shaping for Reinforcement Learning in the Aviation Industry", Master's Thesis, Hamburg University of Technology, October 2022.