No. 85 Physical AI
Tobias Antonsson on the Constraints Shaping Micro-Robotics
The CEO of Bitcraze AB on why micro-autonomous systems are insulated from data center demand, what open hardware needs to be sustainable, and why the physical world remains the most interesting domain.
Conducted in writing · Answers published as received
In this conversation, we spoke with Tobias Antonsson, CEO of Bitcraze AB, about the engineering constraints shaping micro-autonomous systems, the trade-offs behind open hardware, and why low-power physical robotics still matters in an AI era increasingly focused on virtual agents.
This interview was conducted in writing. Answers are published as received.
Currently, the micro-autonomous systems market is somewhat insulated and hasn't been significantly impacted yet. As a result, we are not implementing any unique design considerations beyond our standard practices. However, this situation could evolve in the future as power consumption decreases and the footprint of advanced systems shrinks.
The importance of maintaining flexible production capabilities has increased, not due to specific component-level issues, but because of the need to adapt to trade wars and political implications, allowing for relocation if necessary.
The passive 1-wire system was a simple and elegant solution for deck identification, though it presented limitations. To overcome these, we developed a significantly more dynamic system, at the cost of increased complexity. This new system allows for the creation of more sophisticated decks in the future. The primary drawbacks are higher maintenance requirements, larger PCB real estate usage, and additional manufacturing steps.
The Brushless version is intended to complement the existing ecosystem rather than compete with it, allowing the Crazyflie 2.1+ and the Brushless to coexist. We have focused on streamlining the Brushless experience to make the transition between the two platforms as smooth as possible. The Crazyflie 2.1+ offers the lightest and safest option. While the Brushless version provides significantly more power, it remains very lightweight and minimizes potential harm, especially when equipped with the optional propeller guards.
I don’t see occlusion as a problem in a warehouse-scale Lighthouse v2 deployment. Volume coverage is where things will get hard as the tracking distance of 5-6m from the Lighthouse is a trickier limit. Engineering past that can be done with e.g. a hybrid system together with the UWB-based “Loco” positioning system, preserving precision and determinism while scaling in a controlled way.
Applications that aim for sustained, high-level autonomy expose the limits most clearly. The AI-Deck is well suited for evaluating specific perception or learning tasks, but full, reliable autonomy quickly runs into power and energy constraints at this scale. Today’s more capable edge-AI systems are simply not small or efficient enough yet. That gap is well understood, and it’s exactly what we’re targeting with our next generation of solutions.
As swarms scale, the dominant constraint shifts from individual performance to system-level reliability. Radio bandwidth and synchronization jitter become manageable, but the probability of individual failures inevitably increases with swarm size. At that point, insisting on perfect reliability is counterproductive. The more scalable path is to design systems where occasional individual failures are expected and tolerated, allowing the swarm as a whole to remain robust and effective.
In education, where you want to know all the nitty gritty details of how the system is built, open platforms will always have an edge and will be the preferred and most popular choice. The architectural superiority has to be quite large before it triumphs the openness. So if the regulatory shift pushes towards more openness I think it is just a positive development.
Scenarios where failure results in crashes that could cause harm to property or living things is always the most concerning. That is why we believe small and lightweight solutions are the way to go.
The primary constraint is energy density and the associated expense. It will require a significant period of development and investments for advanced, high-performance systems to become sufficiently miniaturized, cost-effective, and low-powered to achieve practical viability.
A completely open hardware model is difficult to sustain because creating a mutually beneficial, "win-win" environment is highly challenging. It's too easy for others to benefit (to "take") but too hard to get contributions back (to "give").
What we’ve learned is that openness needs structure. Clear boundaries, maintained reference designs, and well-defined contribution paths are what turn an open platform into a sustainable one. That lesson would shape our approach from day one in 2026.
The physical world remains the most interesting and fundamental domain of human experience. While digital spaces grow, the tangible environment is the ultimate stage for human endeavors. We believe technology will continue to enhance this reality, with micro-autonomous systems playing a vital, transformative role.