Yuchi(Allan) Zhao

PhD student @ The Matter Lab UofT | UWaterloo Mechatronics Graduate | Robotics and Computer Vision Researcher

[Yuchi(Allan) Zhao @ZhaoYuchi]

@chris_j_paxton Thanks for your interest of our work! Totally agreed with your point. Being able to focus on local region with VLM and training on a diverse synthetic dataset really make the model not only transfer from sim to real, but also generalize to unseen objects.

[Yuchi(Allan) Zhao @ZhaoYuchi]

@MSewlia @animesh_garg The pre-grasping pose is determined by heuristic. Feel free to let us know if you have more questions.

[Yuchi(Allan) Zhao @ZhaoYuchi]

@MSewlia @animesh_garg Thanks for your interest in our work! We use AnyGrasp for grasping pose prediction and cuRobo for motion planning in NVIDIA IsaacSim. The grasp is triggered when the robot's end-effector reaches its final placement pose, which is calculated based on the output of our model.

[Yuchi(Allan) Zhao @ZhaoYuchi]

Excited to share our latest work on object placement for robot manipulation, AnyPlace! 🚀 A two-stage method that predicts diverse placement poses for objects in the real world for robot to place —trained entirely on synthetic data.

Big thanks to all my collaborators!

[Animesh Garg @animesh_garg]

How can robots reliably place objects in diverse real-world tasks?
🤖🔍 Placement is tough—objects vary in shape and placement modes (such as stacking, hanging, and insertion), making it a challenging problem.
We introduce AnyPlace, a two-stage method trained purely on synthetic data to predict diverse placement poses of unseen objects for real-world tasks.

Read on for more👇

[Animesh Garg @animesh_garg]

How can robots reliably place objects in diverse real-world tasks?
🤖🔍 Placement is tough—objects vary in shape and placement modes (such as stacking, hanging, and insertion), making it a challenging problem.
We introduce AnyPlace, a two-stage method trained purely on synthetic data to predict diverse placement poses of unseen objects for real-world tasks.

Read on for more👇

[Marta Skreta @martoskreto]

Me: "So what are your hobbies?"
ORGANA: "I dabble in chemistry—wherever I go, there’s always a good reaction." 😏

🤖✨ORGANA is on the cover of @Matter_CP this month! 🎉 🧪

Full article: https://t.co/VCrVEb1rPx

Work with awesome collaborators from @acceleration_c 🤗

[Marta Skreta @martoskreto]

Want to escape the freezing winter? 🥶Come join us in Singapore* (*in 3.5 months 😁) at ICLR 2025 for the next instalment of AI4Materials 💪🧪AI4Mat’s first time in Asia and first time at ICLR, wowee!

Submissions due Feb 2. More details here: https://t.co/2FsqCs4RO9

@iclr_conf

[Marta Skreta @martoskreto]

ORGANA and the boyz when mit tech review gives us that shoutout

[Yuchi(Allan) Zhao @ZhaoYuchi]

Super excited to introduce ORGANA, our recent work on a module and user-friendly robotic system that automates chemistry experiments and characterization. Feel free to contact us if you have any questions! Big thanks to my amazing colleagues and advisors!

[Animesh Garg @animesh_garg]

Need a helping hand in the lab? Tired of manually doing tedious chemistry experiments over and over?

Meet ORGANA 🤖 🧪 – a modular and user-friendly robotic lab assistant that can interact, plan and automate a number of chemistry experiments!

[Yi Ru (Helen) Wang @YiruHelenWang]

🤖Excited to present our paper MVTrans: Multiview Perception of Transparent Objects at #ICRA2023!

tl;dr: We introduce a new photo-realistic dataset and multi-task perception network to support end-to-end depth estimation, detection, and segmentation of transparent objects.
🧵1/n

[Florian Shkurti @florian_shkurti]

New paper: how can🤖arms autonomously perform multi-step chemistry experiments in a way that is not scripted & generalizes? We show how task and constrained motion planning can reason about low-level skills to interact with lab instruments and tools.

👇 1/5