Cedric Ith (@cedricith) header
Cedric Ith (@cedricith) profile picture

founding designer @perceptroninc . previously @coinbase

Seattle
Joined May 2013
Photo by Cedric Ith (@cedricith)Photo by Cedric Ith (@cedricith)Photo by Cedric Ith (@cedricith)Photo by Cedric Ith (@cedricith)Photo by Cedric Ith (@cedricith)Photo by Cedric Ith (@cedricith)Photo by Cedric Ith (@cedricith)Photo by Cedric Ith (@cedricith)Photo by Cedric Ith (@cedricith)Photo by Cedric Ith (@cedricith)
Introducing EgoHome: a new open benchmark for dense, sub-second egocentric understanding. Human-annotated household-chore video with 4 annotation types on every clip: hand boxes, subtasks, hand actions, hand captions. It features over 750 segments under 500ms. Data, metrics, and judge prompts are all open. 🧵
25
15
14
123
45,790
we’re opening the waitlist for @mainframe today. it’s a new design tool for designers who ship.
52
18
4
563
29,094
On a different task and different embodiment, we run a similar agent. Here, Mk1.5 sees the room, plans a route, and drives the quadruped accordingly.
1
2
2
25
13,344
Today we're releasing Mk1.5: a new intelligence layer for embodied agents. It flies drones, controls quadrupeds, powers smart glasses, tracks objects, searches the web, reasons visually, and dispatches its own sub-agents. One model, no platform-specific retraining. 🧵
48
57
45
408
147,739
Cedric Ith retweeted
we are hosting a party in our office in Bellevue, WA next Thursday 9/24, featuring live music, food, and drinks!!! RSVP below
1
7
3
18
3,709
Isaac 0.5 fine-tunes to almost any task. New, repetitive tasks like box packing can be done reliably with ~30 episodes. Read more: perceptron.inc/blog/introduc… Weights: huggingface.co/PerceptronAI/…
15
4
98
13,828
We’re making enterprise workloads cheaper and faster. Perceptron Multilook API is now live. Context is prefilled once per call instead of once per question. One video in, up to 16 answers out: 32% of the input cost of sending them separately, and 2× to 4.9× lower median latency under load. Read the report: perceptron.inc/blog/introduc… Create an API Key: platform.perceptron.inc
1
7
4
33
5,176
In case you were wondering – Yes, Isaac can fold tshirts too. More importantly, you can reproduce our results on multiple embodiments. Read more about what Isaac 0.5 can do: perceptron.inc/blog/introduc… Download the weights: huggingface.co/PerceptronAI/…
10
15
2
62
9,969
Isaac 0.5 plays tic tac toe, adapting to small perturbations in game environment (i.e., board placement) and new game strategy (i.e., new opponent). Read more about what Isaac 0.5 can do: perceptron.inc/blog/introduc… Download the weights: huggingface.co/PerceptronAI/…
2
16
65
4,914
Cedric Ith retweeted
I told y'all we were cooking
Here is an unedited version of Isaac 0.5 playing chess, both handling the perception needed for the board, deciding next move (from internal world knowledge or tool call to stockfish), and doing the actuation.
4
14
1,307
Cedric Ith retweeted
I made portals with some open cv and projection mapping :)
154
363
68
5,462
222,169
Cedric Ith retweeted
The best open model for robotics. Let’s go!!!
Today we're releasing Isaac 0.5: 36B dynamic MoE, open weight 🤗 embodied foundation model. Isaac combines multimodal video understanding, embodied reasoning and robot control into a single, sparse backbone.
4
20
3,877
We're releasing open weights, LeRobot training + inference code, evaluation settings and robot-specific configs. We want Isaac to be useful as a base model people can actually adapt and deploy, and we'll continue adding integrations, evals and other assets with the open robotics ecosystem. Reach out if you're working on this :)
1
1
8
173
Here is an unedited version of Isaac 0.5 playing chess, both handling the perception needed for the board, deciding next move (from internal world knowledge or tool call to stockfish), and doing the actuation.
Today we're releasing Isaac 0.5: 36B dynamic MoE, open weight 🤗 embodied foundation model. Isaac combines multimodal video understanding, embodied reasoning and robot control into a single, sparse backbone.
4
8
1
70
8,915
Isaac 0.5: Perceive. Reason. Act.
7
12
6
69
26,594
Cedric Ith retweeted
Inserting frames directly into text flow has been missing from UI design tools. We're fixing that in @paper. Text wrapping included. No more faking that with flex layout. This is still early in development -- tell us how you'd use this and what would make this more useful!
79
29
39
565
94,112
Cedric Ith retweeted
lately, i've been building my own bike app that uses a vision language model to identify different bike parts off an image, and also perform a basic mechanical diagnosis with a DIY repair plan #WIP #bikeSaaS (lol) been super fun to try @perceptroninc's mk1 model🛠️! vlm/ perceptron.inc/blog/introduc… api docs/ docs.perceptron.inc/perceptr…
2
8
1
45
6,909
Cedric Ith retweeted
Combining our detect API frame by frame + BoT-SORT is much more powerful than I thought. One area it falls short is re-identification which our video model will soon aim to solve. docs.perceptron.inc/perceptr… trackers.roboflow.com/develo…
3
4
208
Better data for robot hands! 🤲🏼 Embodied reasoning is the bottleneck now. It’s the data we are missing. Specifically, understanding what the robot's hands are doing throughout a task. @perceptroninc new Egocentric offering tracks both hands through an entire video instead of guessing from sampled frames. That matters because a lot happens between frames. Hand positions shift. Contacts change. The robot adjusts grip. You miss all that if you're sampling. 0.280 semantic end-to-end F1 score on WGO-Bench vs 0.158 from the best pipelines built on Gemini Robotics-ER 1.6. Training a robot to manipulate objects requires understanding every moment of hand-object interaction. If your training data has gaps: missing frames, missed contacts, unclear hand positions, the policy learns from incomplete information. You end up with robots that work in demos but fail in real tasks. Continuous hand tracking gives you the full picture. The policy learns from cleaner, more complete data. That compounds into better real-world performance. ~~ ♻️ Join the weekly robotics newsletter, and never miss any news → ziegler.substack.com
8
42
1
256
18,495
Today we're adding video_frames: send Mk1 the frames you've already decoded on the client, each with its own timestamp. The model sees exactly the frames you send.
1
3
3
17
6,305