Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / The Quiet Crisis in Robot Software
Transcript
- Lucas: There is a massive misconception out there that if you just buy enough robotic arms, the factory runs itself. It is not about the metal anymore. It is about the code that tells the metal where to go. Luna: So the hardware has solved the strength problem, but the brain is still lagging behind? Lucas: Exactly. The hardware is mature. A six-axis articulated arm can move with incredible precision today. But getting that arm to do something useful usually requires weeks of manual programming. That is the hidden cost nobody talks about. Luna: I remember reading about a warehouse project where the robots sat idle for months because the integration team couldn't get the software to talk to the legacy inventory system. Lucas: That is the classic example. You have this expensive piece of machinery sitting on the floor, gathering dust, while engineers spend their days writing spaghetti code to handle edge cases. The marginal cost of adding a new task is too high. Luna: Is that changing with the rise of general-purpose models? Lucas: It is starting to, but slowly. The industry is moving toward what we call sim to real transfer. Instead of teaching the robot in the physical world, which is slow and dangerous, you train it in a perfect digital simulation first. Luna: So you build a digital twin of the entire workspace and let the AI fail millions of times in the virtual world before it ever touches a real part. Lucas: Right. And when those policies are transferred to the physical robot, they work immediately. It cuts deployment time from weeks to days. That is a structural shift in how capital is allocated. Luna: That sounds like the internet was for information. Simulation is becoming the standard interface for physical action. Lucas: It is interesting because the companies winning here are not necessarily the old robot manufacturers. They are the software platforms that provide these simulation environments. They own the data layer now. Luna: So the moat isn't the gear reducer or the servo motor. It is the dataset. Lucas: Precisely. If you have a better simulator, you can generate more training data faster. And if you have more data, your model generalizes better across different factories. It becomes a network effect play. Luna: This reminds me of how cloud computing shifted the burden from hardware maintenance to software updates. Lucas: Spot on. The robot becomes a subscription service of sorts. You pay for the compute power and the model updates rather than just buying the box. The business model is flipping upside down. Luna: Does this mean smaller manufacturers can finally afford automation? Or does the software complexity keep them out? Lucas: It actually lowers the barrier to entry. If the setup is intuitive and relies on natural language prompts or simple visual demonstrations, a small shop can deploy a robot without a dedicated software team. Luna: That would democratize access to high-end manufacturing capabilities significantly. Lucas: It does. But there is a catch. The initial investment in the simulation license and the compute infrastructure is still non-trivial. Not everyone can afford the upfront cost of digitizing their facility. Luna: So we might see a consolidation where larger players buy these software platforms to lock in their suppliers. Lucas: We already are seeing M&A activity focused on robotics software firms. Big tech and big industrial groups are acquiring these startups to secure their future operating systems. Luna: It feels like we are entering the platform era for physical things, similar to what happened with mobile apps. Lucas: That is a fair analogy. The robot is the smartphone, and the simulation environment is the app store. Developers can build specialized skills and sell them to anyone who owns the hardware. Luna: I love that. So the value shifts from the device to the ecosystem. Lucas: Exactly. And this creates a new kind of liquidity in the market. You can buy a pre-trained skill for handling fragile electronics and download it to any compatible arm. Luna: But security must be a nightmare. If a hacker gets into your simulation pipeline, they could tell your robots to destroy themselves. Lucas: Cybersecurity is going to become a primary concern for industrial clients. We are going to see a whole new sector emerging around securing the sim to real pipeline. Luna: It is wild to think that the biggest risk is no longer mechanical failure, but code injection. Lucas: Yes. And insurance companies are going to have to rewrite their policies entirely. They need to understand software risk assessments for physical assets. Luna: I guess we should probably mention that we don't run ads here either, if you want to support that choice directly. Lucas: If today's conversation gave you some clarity on the software side of hardware, consider supporting us through buy me a coffee dot com slash fexingo. It helps keep the lights on and the research deep. Luna: Yeah, we appreciate every bit of support. It lets us dig into topics like this without chasing clicks. Lucas: Back to the main thread though. The other challenge is standardization. Right now, every manufacturer uses a different file format for their simulation data. Luna: So you cannot easily port a model from one vendor's simulator to another's. That fragments the market further. Lucas: It does. We are waiting for an industry-wide standard, similar to usb c for cables. Until then, you are locked into specific ecosystems. Luna: That seems like the next big battleground. Whoever sets the open standard wins the long game. Lucas: Agreed. Open source frameworks are gaining traction among developers, but enterprise adoption requires reliability and support contracts. Luna: So it will likely be a hybrid model. Open core technology backed by commercial support. Lucas: That is exactly what we are seeing. The best platforms allow you to export your models in open formats while keeping their proprietary simulation engines as the premium feature. Luna: It is a clever way to balance community growth with revenue generation. Lucas: Very much so. And it accelerates innovation because developers aren't starting from scratch every time. Luna: Do you think this software revolution will eventually make human operators obsolete on the factory floor? Lucas: I doubt it completely. Humans are still needed for exception handling and creative problem solving. But the role will shift from operator to supervisor. Luna: So fewer hands-on tasks and more monitoring and debugging. Lucas: Exactly. The workforce needs to upskill rapidly. Understanding basic Python or simulation logic will become as common as knowing how to read a blueprint. Luna: That is a significant cultural shift for traditional manufacturing hubs. Lucas: It is. And regions that invest in this training now will have a huge competitive advantage over the next decade. Luna: It makes me wonder if we will see a brain drain from traditional engineering roles into this new robotics software niche. Lucas: Already happening. Salaries for robotics software engineers are outpacing many traditional mechanical roles. The talent war is intense. Luna: So the money follows the code. Which seems inevitable in our current economy. Lucas: It really does. The physical world is being digitized faster than most people realize. Luna: What do you think is the next big leap once we solve the deployment friction? Lucas: I think we will see multi-agent systems. Robots talking to each other in real-time to optimize the entire workflow without central intervention. Luna: Like a swarm intelligence applied to logistics. That sounds incredibly efficient. Lucas: It is. And it removes the single point of failure. If one robot goes down, the others reconfigure instantly. Luna: That level of resilience is what enterprises are really craving right now. Lucas: Absolutely. Supply chains are too fragile. Automation needs to be adaptive, not just rigid and fast. Luna: So the goal is flexibility above all else. Lucas: Yes. Speed is nice, but adaptability is survival. I think the winners will be the ones who prioritize that flexibility in their software architecture. Luna: It is a subtle distinction, but it changes everything about how you design the system. Lucas: It does. And it forces engineers to think more like biologists than machinists. Luna: I like that framing. Evolving systems rather than building static machines. Lucas: Exactly. The factory of the future is alive. It learns and adapts. Luna: And we get to watch it happen.