Online Learning Bridges Performance Gaps Between Identical Robots
Some interesting experiments with our “physical Atari” RL framework — one of our main points was that the flawless simulators didn’t transfer well at all to the messy reality of cameras and servos, but perhaps less obviously, even transferring from one robotroller to a second, theoretically identical, one was a notable performance loss. However, with continual online learning, they recover. High end machinery would be a lot closer robot-to-robot than these 3D printed rigs, but humanoid robots with the stacked joint tolerances probably still have substantial variance.
Weight Decay Softly Prunes Noisy Features, Not Just Simplicity
Weight decay is usually presented as “encouraging simpler solutions”, but I tend to think that the real benefit is the soft pruning of noisy / unhelpful features. Without decay, a weight can random-walk to a large value even if the...
Discover PyTorch’s Pixel_unshuffle: Skip Custom Tensor Hacks
Always a slightly mixed feeling to write pretty good first-principles code to do some tensor rearrangement, only to find that PyTorch has a built in function that does it faster. I had made a point of at least skimming the docs...
Base‑2 Softmax Boosts Fixed‑Point Hardware Efficiency
I wonder if doing softmax in base 2 instead of base e could be a useful optimization on some fixed point hardware. Prepend a 1 to the fractional part and shift left by the integral part as a close-ish approximation...
Consistent Naming Crucial Even With Autocomplete Assistance
It doesn’t matter as much in the age of autocomplete as it used to, but the tiny mental friction of naming inconsistencies like “start_dim” vs “keepdim” still sticks out for me. Carefully pick names as if millions of developers may...
Grok's Blunt Corrections Are Refreshingly Honest
I appreciate how Grok doesn’t sugar coat corrections. https://t.co/b511QzNASc
Sell Remote‑Operated Home Helpers, Not False Robot Dreams
Companies selling the dream of autonomous household humanoid robots today would be better off embracing reality and selling “remote operated household help”. Have teams of employees running them 24/7, with the option to reduce their workload as autonomous behaviors become...