I miss the days when video games were used to showcase technical advances.
"Kates built the game to showcase his additron tube, a miniature version of the vacuum tube, though the transistor overtook it in computer development shortly thereafter."
They sort of still are (Witcher 4 being used to showcase new UE features and software+hardware optimizations is just one example), but we're getting close to the point where we cannot really add more details and realism to video games and they still don't really hammer the hardware. Seems ML is the new showcase if anything :)
> but we're getting close to the point where we cannot really add more details and realism to video games
Lots of human senses aren't tackled by video games yet. Smell, taste, balance, cardioception, proprioception, pain, temperature, pressure are all missing. Where are the immersive tanks or piezzoelectric coveralls that stimulate all of our senses coherently? I bet adding those would hammer the hardware.
this adds a tool to “grade” robot demo episodes by analyzing blur, collisions, and movement smoothness, then filters the bad ones out of the dataset. Seems like a pragmatic way to tackle the data quality problem in robotics would be great to see how much it moves the needle in real training runs.
Training is expensive so I wouldn't necessarily call it "pragmatic".
- the tool's goal is actually to provide a lightweight, practical way to avoid wasting training cycles on bad data.
Evals for robotics are also expensive.
- validation loss is a poor proxy of robot performance because success is underconstrained by imitation learning data
- most robot evals today are either done in sim (which at best serves as a proxy) or by a human scoring success in the real world (which is expensive).
It's great if you have evals and want to backtrack (we're building tools for that too) but you definitely don't want to discover you have bad data after all that effort (learned that the hard way, multiple times).
The metrics the tool scores vary from tedious to impossible for a human to sanity check so there's some non-obvious practical value in automating some of it.
"Kates built the game to showcase his additron tube, a miniature version of the vacuum tube, though the transistor overtook it in computer development shortly thereafter."