У нас вы можете посмотреть бесплатно Multimodality in Robotics. How to predict correct and diverse continuous robot actions или скачать в максимальном доступном качестве, видео которое было загружено на ютуб. Для загрузки выберите вариант из формы ниже:
Если кнопки скачивания не
загрузились
НАЖМИТЕ ЗДЕСЬ или обновите страницу
Если возникают проблемы со скачиванием видео, пожалуйста напишите в поддержку по адресу внизу
страницы.
Спасибо за использование сервиса ClipSaver.ru
In this video, I dive into one of the core challenges in robotics and behavioral cloning: multimodality. This problem shows up whenever there’s more than one correct continuous action for the robot to take next, and classic ML approaches can be confused. Kaggle notebook: https://www.kaggle.com/code/ilialar/m... GitHub repo: https://github.com/IliaLarchenko/robo... I walk through a simple 1D toy task that makes this issue super clear — and then try out different modeling approaches to solve it: 0:00 Intro 2:51 Understanding Multimodal Distributions 7:35 Toy Problem: Robot Obstacle Avoidance 15:15 MSE Regression: Why It Fails 26:30 Switching to MAE Loss 31:24 Alternative Losses Discussion 35:39 Tokenization & Classification Approach 41:52 Probability Sampling Strategy 44:38 Diffusion Policy 1:02:03 Flow Matching 1:11:04 Final Thoughts You’ll see code, charts, and intuitive math for each method. The goal is to build high-level intuition for the problem and popular solutions. This is perfect for ML researchers starting in robotics or anyone curious about the topic. Useful links: ACT: Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware https://arxiv.org/abs/2304.13705 Diffusion Policy: Visuomotor Policy Learning via Action Diffusion https://arxiv.org/abs/2303.04137 Behavior Transformers: Cloning k Modes with One Stone https://arxiv.org/abs/2206.11251 VQ-BET: Behavior Generation with Latent Actions https://arxiv.org/abs/2403.03181 Denoising Diffusion Probabilistic Models https://arxiv.org/abs/2006.11239 DDIM: Denoising Diffusion Implicit Models https://arxiv.org/abs/2010.02502 Flow Matching for Generative Modeling https://arxiv.org/abs/2210.02747 MIT Course on Diffusion & Flow Matching (SDE perspective) https://diffusion.csail.mit.edu/ My DOT-Policy repo https://github.com/IliaLarchenko/dot_... #ai #robotics #deeplearning #machinelearning #statistics #lossfunctions #tokenization #diffusion #flowmatching #multimodality #education