Krishan Rana
@krishanrana.bsky.social
Robot learning and AI researcher focusing on scaling and democratising physically intelligent robots 🌐
https://krishanrana.github.io
https://krishanrana.github.io
Work done at the QUT Centre for Robotics in collaboration with Robert Lee, @halfnelson-au.bsky.social and Niko Suenderhauf.
#robotics #imitationlearning #behaviourcloning
#robotics #imitationlearning #behaviourcloning
June 6, 2025 at 6:06 AM
Work done at the QUT Centre for Robotics in collaboration with Robert Lee, @halfnelson-au.bsky.social and Niko Suenderhauf.
#robotics #imitationlearning #behaviourcloning
#robotics #imitationlearning #behaviourcloning
This work will be presented at the Robotics: Science and Systems (RSS) conference in Los Angeles.
Project Page: imle-policy.github.io
Paper: roboticsproceedings.org/rss21/p158.pdf
Code and Datasets: github.com/krishanrana/...
Project Page: imle-policy.github.io
Paper: roboticsproceedings.org/rss21/p158.pdf
Code and Datasets: github.com/krishanrana/...
June 6, 2025 at 6:06 AM
This work will be presented at the Robotics: Science and Systems (RSS) conference in Los Angeles.
Project Page: imle-policy.github.io
Paper: roboticsproceedings.org/rss21/p158.pdf
Code and Datasets: github.com/krishanrana/...
Project Page: imle-policy.github.io
Paper: roboticsproceedings.org/rss21/p158.pdf
Code and Datasets: github.com/krishanrana/...
IMLE Policy is highly multimodal, and we introduce consistency as a method to ensure consistent mode selection across tasks.
Single-step inference makes it easy to batch generate multiple proposals from which the most consistent with the previous action chunk is selected.
Single-step inference makes it easy to batch generate multiple proposals from which the most consistent with the previous action chunk is selected.
June 6, 2025 at 6:06 AM
IMLE Policy is highly multimodal, and we introduce consistency as a method to ensure consistent mode selection across tasks.
Single-step inference makes it easy to batch generate multiple proposals from which the most consistent with the previous action chunk is selected.
Single-step inference makes it easy to batch generate multiple proposals from which the most consistent with the previous action chunk is selected.
Particularly in the low data regime, we show that IMLE Policy can consistently outperform Diffusion Policy and Flow Matching.
June 6, 2025 at 6:06 AM
Particularly in the low data regime, we show that IMLE Policy can consistently outperform Diffusion Policy and Flow Matching.
We extended this approach to the conditional action generation setting and found that these attributes carried over, enabling us to learn performant visuomotor policies from as little as 17 demonstrations.
June 6, 2025 at 6:06 AM
We extended this approach to the conditional action generation setting and found that these attributes carried over, enabling us to learn performant visuomotor policies from as little as 17 demonstrations.
Implicit Maximum Likelihood Estimation (IMLE) was introduced as a generative modelling algorithm that can surpass Diffusion in the low data regime. RS-IMLE improved upon this to increase the quality of samples.
More details in Ke Li's thread below:
x.com/KL_Div/statu...
More details in Ke Li's thread below:
x.com/KL_Div/statu...
Ke Li 🍁 on X: "Diffusion models turn the data into a mixture of isotropic Gaussians, and so struggle to capture the underlying structure when trained on small datasets. In our new #ECCV2024 paper, we introduce RS-IMLE, a generative model that gets around this issue. Website: https://t.co/WOtUT6lpOq" / X
Diffusion models turn the data into a mixture of isotropic Gaussians, and so struggle to capture the underlying structure when trained on small datasets. In our new #ECCV2024 paper, we introduce RS-IMLE, a generative model that gets around this issue. Website: https://t.co/WOtUT6lpOq
x.com
June 6, 2025 at 6:06 AM
Implicit Maximum Likelihood Estimation (IMLE) was introduced as a generative modelling algorithm that can surpass Diffusion in the low data regime. RS-IMLE improved upon this to increase the quality of samples.
More details in Ke Li's thread below:
x.com/KL_Div/statu...
More details in Ke Li's thread below:
x.com/KL_Div/statu...