Bowen Jiang

PhD Student @UTAustin | Previously MSR student @CarnegieMellon

[Bowen Jiang @bwww08]

Tell a robot “spray the plant” 🪴— how does that translate into dexterous action? It must grasp the bottle stably, aim it at the leaves, and squeeze the trigger.
Excited to share CoDex(#ICRA2026): A framework for dexterous manipulation of functional objects WITHOUT demonstrations using VLM guidance + optimization + RL!

[Bowen Jiang @bwww08]

📄 arXiv paper: https://t.co/lqEmNRc1yO
🔗 Project page & hardware videos: https://t.co/dTMWiLUzkH
Huge thanks to my amazing collaborators @WillReger750 and @RobobertoMM at UT Austin’s RobIn Lab (@texas_robotics)!

[Bowen Jiang @bwww08]

Tell a robot “spray the plant” 🪴— how does that translate into dexterous action? It must grasp the bottle stably, aim it at the leaves, and squeeze the trigger.
Excited to share CoDex(#ICRA2026): A framework for dexterous manipulation of functional objects WITHOUT demonstrations using VLM guidance + optimization + RL!

[Bowen Jiang @bwww08]

💡A key takeaway: analytic grasp candidates often fail during dynamic functional manipulation.
Small in-hand shifts during motion frequently break the precise contact needed for mechanism triggering. Constraint-Guided RL ensures the entire sequence is dynamically robust.

[Bowen Jiang @bwww08]

@meng8131 Thanks!

[Bowen Jiang @bwww08]

Introducing 𝐇𝐀𝐂𝐌𝐚𝐧++, accepted to RSS 2024! We’ve developed a 𝐜𝐨𝐧𝐭𝐚𝐜𝐭-𝐜𝐞𝐧𝐭𝐫𝐢𝐜 action space for RL for manipulation.

Our approach achieves:
•Long-horizon contact reasoning
•Generalization to diverse unseen objects
•Effectiveness across diverse tasks
(1/8)

[Bowen Jiang @bwww08]

@AZabrovskiy Thanks for your interest! We have released our code. Please feel free to check out!

[Bowen Jiang @bwww08]

We have released our code. Feel free to try it on your tasks!

https://t.co/jnF6xN3rLM

Special thanks to @yifengzhu_ut for providing support on Deoxys (https://t.co/GeizpCuwg7) used in our real robot experiments!

[Bowen Jiang @bwww08]

Introducing 𝐇𝐀𝐂𝐌𝐚𝐧++, accepted to RSS 2024! We’ve developed a 𝐜𝐨𝐧𝐭𝐚𝐜𝐭-𝐜𝐞𝐧𝐭𝐫𝐢𝐜 action space for RL for manipulation.

Our approach achieves:
•Long-horizon contact reasoning
•Generalization to diverse unseen objects
•Effectiveness across diverse tasks
(1/8)

[Bowen Jiang @bwww08]

@Stone_Tao Thank you!

[Bowen Jiang @bwww08]

@MRvNone Thanks!

[Bowen Jiang @bwww08]

Curious to learn more? Check out our paper (https://t.co/0b1nREVcOF) and project website (https://t.co/PDCOfceWiQ).

For context, see our previous work HACMan (https://t.co/aACeORRV2B).

Authors: @bwww08 @YilinWu11 @Wenxuan_Zhou @chris_j_paxton @davheld
(8/8)

[Bowen Jiang @bwww08]

One of our key strengths: generalizing to unseen objects. HACMan++ tackles novel items, even in the real world!
(7/8)