Björn Döschl | AI LAB @ UniBw Munich

Björn Döschl

Thu, 12 Feb 2026 00:00:00 +0000

Björn Döschl completed his bachelor’s degree in computer science in Engineering (B.Sc.) in 2020 at the University of Augsburg. Thereafter, he joined the master’s program in Robotics, Cognition, Intelligence (M.Sc.) at the Technical University of Munich. Following a period abroad in Japan with a focus on cyber security, he successfully concluded his master’s thesis in collaboration with the MBDA Deutschland GmbH in November 2023, in the field of situation awareness in multi-agent systems.

Since December 2023, Björn Döschl is a research associate at the Institute of Flight Systems at the University of the Bundeswehr Munich, focusing his research on applying intelligence on unmanned aerial vehicles and building a holistic UAV planning and execution framework, thereby creating a simulated and real-world testing environment. His interests lie in implementing autonomous swarm agents with centralized and decentralized intelligence.

Flight Season 2026 Started

Thu, 12 Feb 2026 00:00:00 +0000

The 2026 flight season has officially started.

During the first test flights of the season, we successfully evaluated the changes made to AUSPEX over the winter on our heterogeneous multi-UAV fleet. The tests allowed us to validate the updated system in real flight conditions and to assess its performance across different UAV platforms.

After several adaptations and refinements, AUSPEX performed very well. These successful initial flights provide a solid foundation for the upcoming experiments and mark a promising start to the new flight season.

AJCAI 2025 @ Canberra

Thu, 04 Dec 2025 00:00:00 +0000

At the 26th Australasian Joint Conference on Artificial Intelligence (AJCAI 2025), held in Canberra, Australia, Björn Döschl presented the paper “Planner-Independent Extraction of Goals and Constraints from Natural Language for Open-World Mobile Robot Missions.” The work introduces a planner-independent pipeline for extracting goals and constraints from natural-language mission instructions and projecting them into planner-specific representations for open-world mobile robot missions.

Conference Details

Conference: AJCAI 2025

Full name: 26th Australasian Joint Conference on Artificial Intelligence

Date: 1–5 December 2025

Presentation date: 4 December 2025

Location: Lowitja O'Donoghue Cultural Centre, Australian National University, Canberra, Australia

Session: Short Oral Session 8: Robotic AI

Summary: Björn Döschl presented research on planner-independent goal and constraint extraction for open-world mobile robot missions.

Paper Details

Title: Planner-Independent Extraction of Goals and Constraints from Natural Language for Open-World Mobile Robot Missions

Authors: Björn Döschl and Jane Jean Kiam

Proceedings: AI 2025: Advances in Artificial Intelligence, Lecture Notes in Computer Science, vol. 16371

Pages: 453–465

DOI: 10.1007/978-981-95-4972-6_35

Springer page: Planner-Independent Extraction of Goals and Constraints from Natural Language for Open-World Mobile Robot Missions

Paper Summary

The paper addresses problem-instance acquisition for robot missions described in natural language. It combines an LLM-based goal and constraint extraction module with a planner-specific projection layer, allowing operator-defined mission instructions to be translated into structured representations for downstream planning systems. The approach is evaluated on mission instructions and demonstrated in AUSPEX for a Search and Rescue UAV scenario.

Planner-Independent Extraction of Goals and Constraints from Natural Language for Open-World Mobile Robot Missions

Mon, 01 Dec 2025 00:00:00 +0000

ICAPS 2025 @ Melbourne

Mon, 10 Nov 2025 00:00:00 +0000

At the 35th International Conference on Automated Planning and Scheduling (ICAPS) 2025, held in Melbourne, Victoria, Australia, members of the AI Lab presented several contributions around UAV planning, simulation, dynamic replanning, and natural-language mission specification.

Max Sommer and Björn Döschl presented the system demonstration “Dynamic REAP: Bringing Life into Simulations for UAV Planning and Acting Frameworks.” Björn Döschl also presented the PlanRob workshop paper “Making Unified Planning Dynamic by Incorporating World State Updates via Feedback Loops” and a poster on “Planner-Independent Extraction of Goals and Constraints from Natural Language for Open-World Mobile Robot Missions.”

Conference Details

Conference: ICAPS 2025

Full name: 35th International Conference on Automated Planning and Scheduling

Date: 9–14 November 2025

Location: Melbourne, Victoria, Australia

Summary: MASC and BJDO presented research contributions on UAV simulation, dynamic planning, and natural-language goal and constraint extraction.

System Demonstration

Title: Dynamic REAP: Bringing Life into Simulations for UAV Planning and Acting Frameworks

Presented by: Max Schnell and Björn Döschl

Authors: Kai Sommer, Björn Döschl, Jane Jean Kiam

Demo page: ICAPS 2025 System Demonstrations

Paper: Download the demo paper

Original PDF: ICAPS demo paper PDF

Dynamic REAP extends UAV simulation for planning and acting frameworks by combining accurate vehicle physics with georeferenced semantic segmentation and dynamic background characters. The system demonstration shows how realistic Search and Rescue scenarios can be used to validate UAV decision-making, perception, and planning pipelines before deployment in real-world missions.

PlanRob Workshop Paper

Workshop: PlanRob: Planning and Robotics

Date: 10 November 2025

Title: Making Unified Planning Dynamic by Incorporating World State Updates via Feedback Loops

Presented by: Björn Döschl

Authors: Björn Döschl, Jane Jean Kiam

This workshop paper presents an architectural and functional extension to the Unified Planning framework for dynamic robot missions. The approach constructs planning problem instances from updated world states, integrates execution feedback, and enables replanning during mission execution. The work is demonstrated in the AUSPEX framework with the REAP simulation environment for UAV-based Search and Rescue missions.

Poster Presentation

Title: Planner-Independent Extraction of Goals and Constraints from Natural Language for Open-World Mobile Robot Missions

Presented by: Björn Döschl

Authors: Björn Döschl, Jane Jean Kiam

Publication: 38th Australasian Joint Conference on Artificial Intelligence

Project page: Publication page

The poster presents a planner-independent pipeline for extracting goals and constraints from natural-language mission instructions. The approach uses an LLM-based goal and constraint extraction module and a planner-specific projection layer, enabling operator-defined mission goals to be translated into structured representations for downstream planning systems.

Presented Contributions

System demonstration: Dynamic REAP for realistic UAV planning and acting simulations.
Workshop paper: Dynamic UP extension with world-state updates and feedback-loop-based replanning.
Poster: Planner-independent goal and constraint extraction from natural-language mobile robot mission instructions.

Gallery

Replace the following placeholders with selected photos from the ICAPS presentations, demo session, workshop talk, and poster session.

Project MENTHON: Workshop at the UniBW

Fri, 31 Oct 2025 00:00:00 +0000

As part of the MENTHON project, a joint demonstration took place on October 31, 2025 together with our partners AID and HAT.tec. We were pleased to welcome representatives from the VDI project management agency as well as associated partners from the DGzRS and the Bergwacht Penzberg.

Technical presentation at the MENTHON workshop
© 2025 AI Lab, UniBw Munich

During the event, we showcased the capabilities of the AUSPEX framework in realistic search and rescue scenarios. One scenario demonstrated how a missing person was located by a heterogeneous drone swarm, identified using advanced object detection, and subsequently supported by a second drone that delivered a first-aid kit directly to the position.

Live demonstration of the drone swarm in action
© 2025 AI Lab, UniBw Munich

In addition to the live demonstrations, technical presentations were held and further real-world mission scenarios were discussed. The day concluded with a joint dinner for all participants. The deployed drone swarm consisted of an MK-20 multicopter (ArduPilot) equipped with a zoom and thermal camera, as well as two Holybro X500 v2 (PX4) drones – one carrying a camera and the other transporting a first-aid kit.

Close-up view of the drone equipment used in the demonstrations
© 2025 AI Lab, UniBw Munich

SaynFly @ IEEE ROMAN 2025

Mon, 25 Aug 2025 00:00:00 +0000

At the IEEE International Conference on Robot and Human Interactive Communication (ROMAN) 2025, held in Eindhoven, Netherlands, Björn Döschl presented his latest research titled “SaynFly: An LLM-Modulo Online Planning Framework to Automate UAV Command and Control.” The conference brought together researchers from robotics, human-robot interaction, and automated systems. SaynFly contributes an LLM-modulo online planning framework for UAV command and control, with a particular focus on search and rescue missions.

Conference Details

Conference: IEEE ROMAN 2025

Date: 25 August 2025

Location: Eindhoven University of Technology, De Zaale, 5612 AZ Eindhoven, Netherlands

Summary: Björn Döschl presented his research on SaynFly at IEEE ROMAN 2025.

Paper Abstract

Command and Control (C2) for Unmanned Aerial Vehicle (UAV) missions requires the dynamic coordination of task planning, environmental perception, and navigation strategies. While Large Language Model (LLM-) modulo planning frameworks have shown promise in addressing such complexity, due to their capabilities to understand natural language and common sense, their limited robustness remains a challenge. Recent adaptions of such frameworks made online planning possible, but they do not scale well with the number of actions. This limits the applicability in robotic frameworks with large sets of parameterized actions, such as those encountered in UAV C2. We introduce SaynFly, an LLM-modulo online planning framework that streamlines the action space by discarding infeasible actions using domain-specific knowledge, while leveraging online heuristic search to mitigate reward uncertainty and automate the C2 processes for UAVs. Test results for our Search and Rescue (SAR) validation scenarios show that SaynFly is 70% more efficient compared to existing frameworks while maintaining or exceeding success rates.

SaynFly: An LLM-Modulo Online Planning Framework to Automate UAV Command and Control

Mon, 25 Aug 2025 00:00:00 +0000

AUSPEX Multi-Drone Success: Heterogeneous Fleet Demonstrates Coordinated Autonomous Operations

Fri, 15 Aug 2025 00:00:00 +0000

We’re thrilled to announce a major milestone for the AUSPEX framework: successful coordinated flight tests with heterogeneous drones running different autopilot systems. Our Multikopter MK-U20 (ArduPilot) and Holybro x500v2 (PX4) both received and executed mission plans from AUSPEX seamlessly.

English

A breakthrough for the AUSPEX framework ( https://github.com/UniBwM-IFS-AILab/AUSPEX ) - we successfully demonstrated coordinated autonomous operations with a heterogeneous drone fleet, strengthening our real-world search and rescue capabilities.

The test involved two different platforms: our Multikopter MK-U20 (ArduPilot) and Holybro x500v2 (PX4). This diversity represents realistic search and rescue scenarios where different organizations deploy various drone types that must work together seamlessly.

AUSPEX generated mission plans distributed to both drones simultaneously. Each platform received waypoint-based search patterns while maintaining coordination with the overall mission. Both drones executed their tasks autonomously, with remote controls ready only for emergency intervention.

The successful integration across different autopilot ecosystems demonstrates AUSPEX’s platform-agnostic design. The framework effectively communicated with both ArduPilot and PX4 systems, proving its versatility for heterogeneous fleet operations - crucial for real-world scenarios where rescue teams coordinate various drone platforms.

This test validates AUSPEX’s ability to handle multi-platform coordination while maintaining the reliability required for critical missions.

German

Ein Durchbruch für das AUSPEX-Framework ( https://github.com/UniBwM-IFS-AILab/AUSPEX ) - wir demonstrierten erfolgreich koordinierte autonome Operationen mit einer heterogenen Drohnenflotte und stärkten damit unsere Fähigkeiten für reale Such- und Rettungsanwendungen.

Der Test umfasste zwei verschiedene Plattformen: unseren Multikopter MK-U20 (ArduPilot) und eine Holybro x500v2 (PX4). Diese Vielfalt repräsentiert realistische Such- und Rettungsszenarien, bei denen verschiedene Organisationen unterschiedliche Drohnentypen einsetzen müssen, die nahtlos zusammenarbeiten sollen.

AUSPEX generierte Missionspläne, die gleichzeitig an beide Drohnen verteilt wurden. Jede Plattform erhielt wegpunktbasierte Suchmuster bei gleichzeitiger Koordination mit dem Gesamtmissionsziel. Beide Drohnen führten ihre Aufgaben autonom aus, wobei die Fernbedienungen nur für Notfälle bereitgehalten wurden.

Die erfolgreiche Integration verschiedener Autopilot-Ökosysteme demonstriert das plattformunabhängige Design von AUSPEX. Das Framework kommunizierte effektiv mit ArduPilot- und PX4-Systemen und bewies seine Vielseitigkeit für heterogene Flottenoperationen - entscheidend für reale Szenarien, bei denen Rettungsteams verschiedene Drohnenplattformen koordinieren müssen.

Dieser Test validiert die Fähigkeit von AUSPEX, Multi-Plattform-Koordination zu bewältigen und dabei die für kritische Missionen erforderliche Zuverlässigkeit zu gewährleisten.

#UAV #DroneSwarm #AUSPEX #SearchAndRescue #ArduPilot #PX4 #HeterogeneousFleet #AutonomousCoordination

AUSPEX in SAR Operations: An Integrated Open-Source Decision-Making Framework for UAVs in Rescue Missions

Tue, 25 Mar 2025 00:00:00 +0000

LEAP @ CoRL 2024

Sat, 09 Nov 2024 00:00:00 +0000

At the LEAP (Learning Effective Abstractions for Planning) workshop, held as part of the Conference on Robot Learning (CoRL) 2024, Björn Döschl presented his latest research titled “Say-REAPEx: An LLM-Modulo UAV Online Planning Framework for Search and Rescue.” The event brought together leading researchers in planning and robot learning, with a focus on abstraction-based methods. The workshop fostered engaging discussions and offered valuable feedback from peers working at the forefront of AI and robotics.

LEAP @ CoRL 2024

Sat, 09 Nov 2024 00:00:00 +0000

Workshop Details

Event: LEAP Workshop @ CoRL 2024

Date: 9 November 2024

Location: Science Congress Center Munich, Walther-von-Dyck Str. 10, 85748 Munich, Germany

Summary: Björn Döschl presented his research at the LEAP workshop during CoRL 2024.

Event Abstract

The Learning Effective Abstractions for Planning (LEAP) workshop, part of the Conference on Robot Learning (CoRL), took place on 9 November 2024 at the Science Congress Center Munich. Björn Döschl presented his work titled "Say-REAPEx: An LLM-Modulo UAV Online Planning Framework for Search and Rescue" in the form of a poster and gained valuable insights into cutting-edge research in the field of abstractions for robot learning and planning.

Say-REAPEx: An LLM-Modulo UAV Online Planning Framework for Search and Rescue

Fri, 25 Oct 2024 00:00:00 +0000