Dongjie Yu

🧊 Ablation We ablate key design choices of ZPRL, including whether the bottleneck is used, the perturbation scale λ, the dimension of z, and the offline data size. Here is the takeaway: 1. A compact, task-relevant bottleneck latent matters. 2. λ must be large enough to steer, but small enough to stay local. 3. More offline data gives RL a better latent manifold to exploit.

Great thanks to @kunlei15 , @ZhennanJia13007 , Prof. Jia Pan, @HarryXu12 , @ChiChu299792458 , @TianRainWind for contribution, discussion and supervision during this work!

🦾 Introducing ZPRL: RL steers pretrained manipulation policies through a compact bottleneck latent, with smooth and efficienct exploration. More can be found in: Website: manutdmoon.github.io/ZPRL/ Code: github.com/manutdmoon/ZPRL Paper: arxiv.org/abs/2605.19919

🤔🤖Robots can throw, but can they play ring tossing like humans? 🦾🎯Introducing 𝐃𝐀-𝐌𝐌𝐏 — a dynamics-aware motion manifold primitives framework for learning coordinated and accurate throwing.

A ~35minute talk I gave at the ICRA 2022 Behavior Priors in Reinforcement Learning for Robotics Workshop youtube.com/watch?v=SRV_X4…