I recently attended NVIDIA's GTC, SPIE Photonics, a French SF meetup, and Designcon, and I wanted to share my key takeaways regarding the world of robotics.
One thing that stood out was the sheer number of humanoid robot demonstrations – far more than I've seen before. These impressive machines can now run for several hours on battery power. They're typically built strong, using metal alloys, and can handle a good amount of weight. Interestingly, some companies, like Fraunhofer, are creating designs that drift away from humanoids that are more affordable due to less freedom of movement and can carry even heavier loads.
If you're in the industrial automation space, SPIE is the place to connect with part manufacturers. You can find a wide range of precision industrial robots, generally priced between $4,000 and $30,000. Many are surprisingly easy to program for simple tasks, and vendors often provide helpful free services.
The French independent ecosystem is always interesting, where the real engineering challenges affect BOM pricing more than power politics. They also give good example, how to build sophisticated systems in aerospace, electronics, or industrial automation with limited resources.
AI is proving to be a gamechanger, significantly lowering the barrier to entry for robotics startups. Development costs have plummeted thanks to AI, and we're even seeing the emergence of text-to-motion models. Traditionally, industrial robotics projects could take six months and cost a million dollars to complete. The rise of AI, TPUs, and generative motion models promises to dramatically reduce these figures.
NVIDIA offers a wealth of tools and libraries for robotics through its Jetson family. However, their solutions tend to be on the pricier side, especially if you're planning to analyze camera sensor data.
China has become a major force in robotics, producing hundreds of thousands of qualified STEM graduates each year. Their funding landscape relies heavily on private equity, and state-driven conglomerates are often able to compete, collaborate and acquire successful startups at a premium.
Despite the advancements, widespread robotics adoption still faces hurdles. Safety and a lack of clear regulations remain key concerns. Many Western companies have used robots for mass manufacturing for decades, but deploying automated manufacturing for new products at scale can raise defense-related issues. Many companies rely on experts with decades of experience, and there's a significant shortage of similarly skilled workers.
The traditional reliance on ARM, DSPs, and NVIDIA Jetson devices often comes with strict NDAs and high costs. Sensors are prone to drift and divergence, requiring frequent calibration. This makes maintenance a similar challenge to that of cars and commercial trucks. Since robots compete for the same funding and financing, these warranty and lifetime issues are limiting factors for large-scale adoption. Startups can quickly create a working prototype with AI, but developing subsequent versions requires calibration and skilled labor.
One interesting example I saw involved retrofitting existing excavators with industrial robots. This approach reduces development costs by leveraging existing products and expands the user base for robotic upgrades. The cost of automation can be as little as 10% of the $500,000 price tag of a new excavator, significantly reducing the bill of materials.
Finally, I see a huge opportunity in using humanoids to automate existing cars and trucks. Competing with existing human labor can establish the right market prices and create a strong, predictable market that's attractive to investors.