> Currently I’m thinking that cameras connected to deep learning networks is the technology with the most promise to enable a new wave of low-cost robotic applications.
After his earlier comments about fancy sensors being too expensive I was surprised to read this. But I agree. Once some cheap neural net ASICs come out I expect a revolution in robotics will follow soon after.
I'm not sure if I understand your main thrust, but running data through a trained NN can be done in relatively few operations - it is training the network which takes a lot of compute resources.
Cameras are pretty cheap and already mass produced like crazy. The expensive part is the processing you need to extract high level information from the pixels. That's why we need cheap neural net ASICs.
After his earlier comments about fancy sensors being too expensive I was surprised to read this. But I agree. Once some cheap neural net ASICs come out I expect a revolution in robotics will follow soon after.