- Simulation interoperabilityLead author of REP 158, the vendor-neutral OpenUSD standard for interoperable robot simulation assets, accepted by the ROS and Gazebo committees in 2026
- Adopted by industryProposed and largely wrote the ROS Simulation Interfaces, now in key simulators: Gazebo, NVIDIA Isaac Sim and O3DE
- Recognised expertExpert reviewer for the European Commission since 2018; main-stage keynote, Davos Tech Summit 2026; talks, workshops and demos at ROSCon and ROS World since 2016
- Active mentorFounded and leads the ROS Embodied AI Community Group; brings AI coding agents to Robotec.ai's engineers; volunteer lecturer at the University of Warsaw
- Slingshot: new productDevised and architected an end-to-end flow with high-fidelity simulation for physical AI, and wrote most of the code: a walk-through of a real site becomes a simulation-ready digital twin in days, not months
Adam's projects and initiatives
Key architect and software engineer of the following software projects (internal and open-source), and lead author of the standards among them: from 2018 to 2026, with his role in each and where each stands today.
ROS: Robot Operating System, the most widely used open-source software framework for robots. REP: Robotics Enhancement Proposal, the standards process of the ROS and Gazebo projects. OpenUSD: Universal Scene Description, the open 3D scene format used across film, games and simulation. O3DE: Open 3D Engine. Gazebo and NVIDIA Isaac Sim: widely used robot simulators. PMC: project management committee.
| Year | Project | His role | Status today |
|---|---|---|---|
| 2026 | open standardREP 158: OpenUSD conventions for robot simulation assets | Lead author (with co-authors from Robotec.ai, NVIDIA and Ekumen); submitted the core schema; sought feedback from MuJoCo | Accepted by the ROS and Gazebo project management committees, September 2026; a public compliance checker (ros-usd-check), built by the Robotec.ai team |
| 2026 | internal productSlingshot, a Robotec.ai product: a walk-through recording of a real site becomes a simulation-ready digital twin in days, not months | Conceived and architected it, and wrote most of the code: LiDAR-grounded 3D reconstruction (Gaussian splatting) trained region by region across GPUs, self-tested correction of camera calibration and timing errors, removal of people for privacy, object reconstruction, a native Rust core and OpenUSD export; built with an AI coding-agent harness he designed | In Robotec.ai's product line; its OpenUSD output is checked against REP 158 and opens in NVIDIA Isaac Sim and O3DE |
| 2024 | open standardROS Simulation Interfaces: a common ROS 2 interface for robot simulators | Proposed it (REP issue 410); wrote the first version (pull request #1) and most of today's interface definitions; developed it with NVIDIA contributors and reviews from Gazebo and ROS maintainers; listed maintainer | Implemented in Gazebo, NVIDIA Isaac Sim and O3DE; released for ROS 2 Humble to Rolling; a dependency of 60+ public GitHub repositories, among them MVSim and bridges for Genesis and Newton |
| 2024 | open sourceRAI: open-source agent framework for embodied AI | Conceived and directed it; co-author of the PAAMS 2025 paper | Developed by the RAI team at Robotec.ai; cited in later research, for example RHO (2026), which evaluates on RAI's O3DE benchmark (two RHO authors co-presented a ROSCon 2026 workshop with him), and Embodied Playground (2026) |
| 2022 | open sourceO3DE ROS 2 Gem: robotics in Open 3D Engine | RFC, first code; top committer of the original repository | The official O3DE robotics stack, maintained by O3DE SIG-Simulation; he was its Chair (elected 2023) |
| 2020–2021 | open sourceROS 2 for Unity and ros2cs | Co-creator; first commit in both, top committer of ROS 2 for Unity; architected and engineered Robotec.ai's first work on AWSIM | Ships inside AWSIM, an open-source driving simulator for Autoware started by TIER IV; for the Indy Autonomous Challenge teams' simulator, “@adamdbrw's ros2-for-unity gives a very large performance improvement” (Joe Speed) |
| 2020–2021 | open sourcerosbag2 performance: recording robot data | Benchmarks and report, optimised SQLite defaults, --regex and --exclude; presented the findings to the ROS 2 TSC | Shipped in ROS 2; the optimised SQLite defaults are still documented; ROS 2 Galactic release notes: in a 200 Mbps stress test Foxy dropped up to 70% of messages, Galactic kept about 100% |
| 2018 | open sourcemicro-ROS (EU project OFERA) | First author of the founding scenarios and requirements (cited in the micro-ROS chapter of Robot Operating System: The Complete Reference) | ROS 2 on microcontrollers; a Zephyr Project module |
Business impact: technology vision, strategy and product
From one of Robotec.ai's first five people to CTO: what his technology leadership has delivered for the company and for ROS 2 users.
- Company growth. Among the first five people at Robotec.ai, he has led its technology through bootstrapped growth to 50 people, building simulation and digital-twin products and services for customers such as Boliden, Qualcomm, Analog Devices and TIER IV.
- Strategy and roadmap. As CTO he owns Robotec.ai's technology strategy and roadmap: open standards and open-source collaboration as the base, and a path from services to products (the RoSi digital-twin platform, Slingshot). The two meet in Slingshot, whose OpenUSD output is checked against REP 158, of which he is lead author.
- Product. He conceived Slingshot, a new Robotec.ai product, designed it and wrote most of its code. It turns a site walk-through into a simulation-ready digital twin in days, not months.
- Reliability at scale. Led the rosbag2 performance work behind ROS 2 Galactic's recording gains: in a 200 Mbps stress test, Foxy dropped up to 70% of messages and Galactic kept about 100%. Apex.AI, ADLINK, Bosch and TIER IV funded and supported the work.
- Market position. His standards work is credited publicly to Robotec.ai. NVIDIA, on the ROS Simulation Interfaces: “Spearheaded by Robotec.ai”; the Open Source Robotics Alliance: “Robotec.ai: Leading the development of new standardized interfaces for simulators”.
Customer by customer: customer success stories below.
Customer success stories
Public case studies and statements from Robotec.ai's customers, partners, sponsors and open-source users, among them Boliden, Qualcomm, Analog Devices, AWS, TIER IV, NVIDIA and Dexory, with his part in each.
| Customer | Outcome | His role | Source |
|---|---|---|---|
| Boliden | Safe validation of autonomous mining fleets before deployment, in a digital twin of a large mining operation that replays recorded system data and supports predictive analytics. | As CTO, directed the engineering. | case study |
| Boliden | A UWB safety system's geofences and collision risks verified before go-live, in Kite Shield: a high-fidelity digital twin and mixed-reality environment of Boliden's underground tunnels and fleet. Kiteshield, funded by Boliden and LKAB, lists Robotec as a partner for simulation and digital twins. | As CTO, directed the engineering (RoSi platform). | case study |
| TIER IV | AWSIM, the open-source driving simulator for Autoware, “developed by TIER IV and Robotec.ai” (Indy Autonomous Challenge), runs on ROS 2 for Unity and Robotec's GPU lidar simulation. TIER IV's simulation engineer: it lets them “reproduce simulations that are as close to real cars as possible” (Unity case study). | Architected and engineered Robotec.ai's first AWSIM work (2022) and co-created ROS 2 for Unity (2021), the communication layer AWSIM runs on; as CTO, directs the current work with TIER IV. | case study |
| Qualcomm | Real-time performance confirmed on Qualcomm edge hardware, through synthetic data and hardware-in-the-loop validation in a photorealistic retail simulation of a mobile manipulator restocking shelves. | As CTO, directed the engineering. | case study |
| Analog Devices | Digital twin of a semiconductor production bay to validate robotic workflows, human–robot interaction and automation efficiency. | As CTO, directed the engineering (RoSi platform). | case study |
| AWS | Cloud-native, high-fidelity robotics simulation at scale with Open 3D Engine and ROS 2: multi-robot scenarios and distributed sensor simulation, on the ROS 2 Gem “developed by Robotec.ai and AWS Game Tech” (Husarion). | Built on the O3DE robotics stack he started; he was VP Robotics & Simulation. | case study |
| NVIDIA and Intrinsic | Now in Gazebo, NVIDIA Isaac Sim and O3DE, and in the AI for Industry Challenge toolkit: a standard ROS 2 interface for controlling robot simulators, for interoperability across them. NVIDIA: “Spearheaded by Robotec.ai”. | Proposed the standard and is its lead author. | case study |
| Apex.AI, ADLINK, Bosch, TIER IV | 6x faster robot data recording for the Indy Autonomous Challenge: “Robotec.ai rewrote rosbag2 making it 6x faster with funding & engineering support from Apex.AI, ADLINK, Bosch, Tier IV”, shipped in ROS 2 Galactic. | Led the rosbag2 performance work and wrote its report. | Joe Speed, ROS Discourse |
| Dexory | From one library to the full stack: Dexory adopted Robotec.ai's GPU lidar simulation (RGL) for its warehouse robots in 2023, when Guillaume Doisy, its Lead Systems Architect, reported “Working now!”, kept its fork current into 2026 and contributed features back; it now runs on Robotec.ai's simulation stack. | Directed and co-built the simulation stack Dexory now runs on, and directed the engineering behind RGL, as VP and then CTO. | Dexory's fork; RGL credits |
Robotec.ai's open-source simulation libraries are also used by Autoware's AWSIM, Japan's national 3D city-model programme Project PLATEAU (lidar simulation) and the OpenLoong humanoid project (ROS 2 for Unity).
Mentorship and teaching
He mentors technical leads and engineers at Robotec.ai, lectures to students and brings developers together in the wider robotics community.
- Mentors Robotec.ai's technical leads in understanding customers' business needs, project management and leadership.
- Mentors and develops Robotec.ai's engineers, including junior engineers, and drives the company's adoption of AI coding agents with the harness he built for Slingshot. He also helped Robotec.ai's non-technical teams adopt agentic workflows.
- Founded and leads the ROS Embodied AI Community Group (since 2024), which brings together developers applying AI agents and generative AI to robots.
- Volunteer guest lecturer at the University of Warsaw, his alma mater: lectures at the Faculty of Mathematics, Informatics and Mechanics, and a lecture on large language models for linguists.
- Robotec.ai's lead for a University of Warsaw student team project on large language models in robotics (2023/24), with day-to-day mentoring by Mateusz Żak.
- Led teams: PIAP's Autonomy group (11 people) and, as VP, Robotec.ai's Robotics and Simulation department of six teams.
- Volunteer maths teacher for children from families in hardship (2009–2010).
Publications
- Rachwał, K., Majek, M., Boczek, B., Matejczyk, J., Matejkowski, D., Dąbrowski, A., Seyde, T., Amini, A., Ganzha, M. Multi-Agent Robotic Control with Onboard Vision-Language Models. Advances in Practical Applications of Agentic AI and Multi-Agent Systems: The PAAMS Collection (PAAMS 2026), Springer LNCS, pp. 427–432.
- Rachwał, K., Majek, M., Boczek, B., Dąbrowski, K., Liberadzki, P., Dąbrowski, A., Ganzha, M. RAI: Flexible Agent Framework for Embodied AI. Highlights in Practical Applications of Agents, Multi-Agent Systems and Computational Social Science: The PAAMS Collection (PAAMS 2025), Springer CCIS, pp. 195–206.
- Report on performance improvements in rosbag2. ros2/rosbag2 technical report, 2021.
- Maciaś, M., Dąbrowski, A., Fraś, J., Karczewski, M., Puchalski, S., Tabaka, S., Jaroszek, P. Measuring Performance in Robotic Teleoperation Tasks with Virtual Reality Headgear. AUTOMATION 2019, Springer, pp. 408–417.
- Czarnowski, J., Dąbrowski, A., Maciaś, M., Główka, J., Wrona, J. Technology gaps in Human-Machine Interfaces for autonomous construction robots. Automation in Construction 94 (2018), pp. 179–190.
- Dąbrowski, A., Lütkebohle, I., Lange, R., Losa, J. M., Gamarra, B. O., Goenaga, I. M., Vilches, V. M. Reference Scenarios and Technical System Requirements Definition. OFERA (Horizon 2020) deliverable, 2018. Cited in the micro-ROS chapter of Robot Operating System: The Complete Reference, vol. 7.
- Identification and analysis of gaps in the field of UGV human-machine interfaces. NATO AVT-241 Specialists Meeting, 2015. With M. Maciaś and J. Główka.
Also on ORCID.
Point of view
Physical AI will be won by engineering, not only by bigger models: safety that holds “even with the dumbest AI in charge”, simulation that tests operations at scale before anything touches the floor, modular robot skills instead of monoliths, and small, additive wins first.
Today's demos deliver about 90%. The industry runs at 99.9%.
From The Blueprint for Physical AI, his keynote in the main-stage session Foundation Models in Robotics, Davos Tech Summit, July 2026 (slides, PDF)
Career
- Since 2024
- CTO, Robotec.ai, Warsaw. Technology vision, strategy and engineering across robot simulation, digital twins, physical AI and open source. Among the first five people at Robotec.ai, he has led its technology through its bootstrapped growth to 50 people.
- 2017–2024
- Robotec.ai (founded 2019; 2017–2019 at Robocar, its predecessor under the same CEO): Technical Lead (2017), Principal Software Engineer (2021), VP Robotics & Simulation (2023), leading a department of six teams.
- 2014–2019
- PIAP, Industrial Research Institute for Automation and Measurements (2017–2019 half-time, alongside Robocar). Led the Autonomy group (11 people, robotics and deep learning) and PIAP's part in two EU projects: OFERA, which produced micro-ROS, and R5-COP (30 partners, modular ROS-based robots), whose exhibit won the ARTEMIS Exhibition Award in 2015.
- 2005–2014
- Software engineer and system analyst: work on the C++/Qt client of Gadu-Gadu, then Poland's most-used instant messenger; system analyst on integrating information systems at PZU, Poland's largest insurer; digital TV decoding; a risk-rating system for Poland's largest bank at Deloitte.
- Since 2018
- Independent Expert Evaluator, European Commission (Horizon 2020, Horizon Europe): reviews research and innovation proposals for architecture, feasibility, execution risk and consortium capability.
- Education
- MSc in Information Technology, University of Warsaw. Polish (native), English, German.
Full career, publications and roles: printable profile.
Selected talks
| Date | Event | Title |
|---|---|---|
| Sep 2026 | ROSCon, Toronto | workshopTrain, Simulate, Deploy: Agentic AI from Cloud to Robot, co-presenter |
| Jul 2026 | Davos Tech Summit | keynoteThe Blueprint for Physical AI, main stage |
| Oct 2025 | ROSCon, Singapore | talkIntroducing the new ROS simulation standard (video), with Ayush Ghosh (NVIDIA) |
| Oct 2024 | ROSCon, Odense | talkAgent-based AI Framework for ROS 2 (video) |
| Oct 2023 | ROSCon, New Orleans | talkSimulate robots like never before with Open 3D Engine (video) |
| Oct 2016 | ROSCon, Seoul | talkEvaluating the resilience of ROS2 communication layer (video), with Rafał Kozik and Mateusz Maciaś (PIAP) |
All talks, demos, panels and podcasts: speaker and press kit.
Community and standards roles
- Leader of WG5, Simulations & Standards, and Management Committee member for Poland, COST Action Agentic Physical AI (API-Net, CA25178), a European research network (since 2026)
- Independent Expert Evaluator, European Commission (since 2018)
- Founder and lead, ROS Embodied AI Community Group (since 2024)
- Formerly Chair (elected 2023) and maintainer, O3DE SIG-Simulation
- Initial member of the Physical AI SIG of the Open Source Robotics Alliance (OSRA) in 2025; coordinated its Embodied Agentic AI working group when it began
What others say
“Led by Robotec.ai and developed in collaboration with Gazebo, Open 3D Engine, and NVIDIA, these interfaces streamline integration by offering developers a consistent and simplified way to manage simulations directly from ROS 2.”
“Thanks @adamdbrw for kicking this off. I do agree that having a standardized simulation interface would be good for the community.”
“Standard ROS interfaces have been defined for interacting with simulation in the ros-simulation/simulation_interfaces package, which has just had its 1.0.0 release by @adamdbrw. We should consider how to support these interfaces in Gazebo.”
More statements, each linked to its source: Sources.