Interlatent

Hardware is the bridge between AI and the physical world Atoms and bits must work together to create future systems embedded with physical intelligence We wrote a guide for those curious about the atoms.

@z_aaa_th i like this color

@interlatent My SO-101s approve this message

Hardware is the bridge between AI and the physical world Atoms and bits must work together to create future systems embedded with physical intelligence We wrote a guide for those curious about the atoms.

@ddanielshin for real man

@interlatent very useful

@roland_graser thank you chief deep tech officer

@interlatent siiickk!!

This post was done in collaboration with @tnkrdotai

link: interlatent.com/blog/interlate…

@bevelez will keep that in mind for the future, thank you.

@interlatent I think this was a good read, I think if I was going to give you 1 piece of feedback to make it stronger, it would be making the opening section break down what the article will teach someone, what concepts a reader will take away, and then a table of contents or deep-linking.

Our mission is to make it easy for anyone to deploy a robot to help them in the real world We wrote an intuitive guide to understanding modern robotics, catered toward an audience that understands technology but not AI robotics We hope that this short blog post embeds in you the core principles that will bring further curiosity.

@z_aaa_th Thank you! The visualization is definitely exaggerated for the purpose of understanding But you can imagine that when a robot moves in chunks it controls the trajectory better because there are less opportunities of producing an out of distribution state

@interlatent Great overview of the current state of robotics! Here's a gap that I ran into: Compounding error in the discrete (non-chunk) action visualization looks exaggerated. How does ACT approach combat the same issue, compounding error between chunks?

@BangMingYong no clever tricks, same tools as most other websites

@interlatent This is a very beautiful website. How are the animations created??

@nikhilkr yes on paper everything sounds easy in fact, everything is a function on paper

the policy-as-function framing is the right starting point the part that takes years to really feel is the observation space is never clean in the real world. camera pixels get occluded, joint angles drift, gripper force readings lie. you spend months making the inputs trustworthy before the policy has any chance of being good

@roland_graser excited!

@interlatent awesome!!

there are also many diagrams you can play around with. thank you @ddanielshin

@0xyeung123 means a lot

@interlatent Learned a lot

@leylndd thank you!

@interlatent nice work

link: interlatent.com/blog/interlate…

We propose a reward shaping mechanism which aims to reward positive parts of rollouts while adding time step penalty to discourage idleness or inefficiency. More in the blog post below interlatent.com/blog/automatic…

Introducing our first technical blog post: Automatic Dense Rewards for Autonomous Robot Learning In this post, we utilize VLM chunking techniques to densely reward trajectories based on video and telemetry data without human supervision.