Robot Utility Models (RUMs) enable basic tasks โ door opening, drawer opening, object reorientation, etc. โ at ~90% accuracy without ANY finetuning (i.e. zero-shot) in unseen new environments.
Fully open source!!! models, data, code & hw.
We think this is super exciting, why?๐
1. Unlocks many practical home utility tasks that often involve these basic tasks as part of an action chain. โGo get me a forkโ involves opening the kitchen door and then opening the cutlery drawer.
2. This works well **zero-shot in unseen and new** environments, which is practically a huge deal. Turn the robot on, and get going.
3. The recipe for building a new model is fairly generic, and we think with a bit more refinement this can be a general recipe to build many more Utility models.
More details and access ๐
@francoisfleuret I've only just now come across your generated culture work, having read the PDF you've linked on your website. I would assume more models would decrease the odds of this trick succeeding, is there a reason there are 3 models now instead of the 5 in Self-Generated Culture?
Y'all better hop on over to Puzzmo for today's daily crossword, written by yours truly! I had tons of fun making it and hopefully you'll enjoy it! It's short and sweet!
I make wacky local multiplayer experiences with midi controllers that are installed in spaces. This one has the crowd controlling visuals and parameters of the game using a midi controller while 4 players battle in a glitchy mess of audio reactive visuals!
#gamedev#unity
I made a 4-? player chaotic local multiplayer battle experience built inside of an audio reactive visualizer. Four players battle while the crowd plays with a grid of 16 knobs and buttons that each affect visuals and parameters of the game.
#indiegames#gamedev#unity
@simonw I named the script `stt` and this makes it so I don't have to remember specific arguments (the int8 flag allows it to run on apple silicon, won't work otherwise), simply `stt file.mkv` and then let it run.
@simonw I can rec WhisperX, found it when looking for a solution with builtin diarization (speaker recognition), and it's generally *really* fast. I use it via this shell script: /opt/homebrew/Caskroom/miniconda/base/envs/whisperx/bin/whisperx --model large-v3 --compute_type int8 "$@"
@simonw Hey Simon, out of curiosity, how do you decide which quantized version of llama 3 to use? I noticed you suggested Q4 on your blog and was wondering what tradeoffs you've noticed if any between it and the versions on either side.