Tianyu Li EasyPaperSniper
@SniperPaper
Followers
169
Following
137
Media
12
Statuses
40
CS Phd@Gatech | exResearch Scientist@FAIR
Joined September 2019
⭐️ Takeaway: Commercial game consoles aren’t just entertainment. They can be standardized, physically grounded benchmarks for Embodied AI. A more thorough version of this work is coming soon—stay tuned!. #JustDance #Unitree #EmbodiedIntelligence #AIResearch.
0
0
1
Here’s an example validation on Unitree G1 humanoid (averaged over 3 2-star level dances from Just Dance):.🤖 Robot avg. score: 5,707.🧑 Human avg. score: 9,361.Not the upper bound—future engineering can push robots much closer to matching human performance.
1
0
0
Our pipeline is built on top of open-source projects 🙌:. Motion capture with ROMP: Motion retargeting with GMR: Whole-body control with GMT: Plus Joy-Cons for scoring 🎮
1
0
0
The Nintendo Switch is more than just fun 🎮—it can also advance humanoid research! @sehoonha @jeonghwankim0 @wontaek0820 @donghoonbaek @seungeun071 . With Switch4EAI, we turn Just Dance into a benchmark for humanoid whole-body motion tracking:.✅ ~$400 teleoperation setup.✅
1
1
6
RT @ICatGT: Wrapping up @GeorgiaTech’s technical tours today, hosting visitors from @icra, with a sword wielding spot demo from @sehoonha a….
0
2
0
100% agree. Our focus should be on 'intelligence' rather than 'humanoid,' which is just one platform for measuring intelligence.
Embodied intelligence is more than just humanoid robots—it’s a broader paradigm. From self driving cars to smart environments, it’s how embodied systems spatially reason, learn, interact, and evolve in the physical world. Many researchers and VCs are missing the point by.
0
0
3
For more information, please visit our paper:and project website: This work is done during my internship at FAIR, Meta @AIatMeta . I want to thank all my great collaborators: @xavierpuigf @sehoonha @joannetruong Jimmy Yang, Akshara.
arxiv.org
Moving large objects, such as furniture or appliances, is a critical capability for robots operating in human environments. This task presents unique challenges, including whole-body coordination...
0
0
2
By combining our trained object manipulation policy with a heuristic planner, we demonstrate an automated dining chair rearrangement application. (7/8)
1
1
2
The trained policy can be integrated with a high-level planner for real-world applications. By combining our object manipulation policy with user commands, we demonstrate its effectiveness in real-world scenarios—such as moving large trash carts. (6/8)
1
1
0
RobotMover enables stable manipulation, while baseline methods struggle with objects falling off or colliding with the robot. (5/8)
1
0
0
Our results show that the learned policy can move objects in different directions and is robust enough to handle objects of various materials and shapes. (4/8)
1
0
1
Human demonstrations offer valuable guidance for these tasks, but how can we imitate them when robots and humans have different morphologies? 📷.RobotMover introduces the Dynamic Chain, a graphical representation of agent-object interactions. By leveraging the Dynamic Chain,
1
0
0
Moving large objects is challenging. Their size increases the risk of collisions, potentially damaging both the object and the robot. Additionally, their greater weight can cause instability, making them prone to slipping from the robot's grasp due to underactuated dynamics.
1
0
2
We've seen robots move like our favorite athletes. We've watched them fold clothes and do the dishes. Now, it's time for robots to help you move furniture. Introducing RobotMover—a learning framework that enables robots to acquire object-moving skills from human demonstrations.
1
5
15
RT @ChongZitaZhang: This is the mobile manipulation I want to see. You can only get this via RL.
0
31
0
RT @UnitreeRobotics: Unitree B2-W Talent Awakening! 🥳.One year after mass production kicked off, Unitree’s B2-W Industrial Wheel has been u….
0
4K
0
I've moved all my current projects to Genesis. Genesis is fast, physics realistic and easy to use. 1000% worth a 'pip install' if you are working on Robotics, Embodied AI and all related topics.
Everything you love about generative models — now powered by real physics!. Announcing the Genesis project — after a 24-month large-scale research collaboration involving over 20 research labs — a generative physics engine able to generate 4D dynamical worlds powered by a physics
4
6
71
RT @xwang_lk: It is just so sad that the #NeurIPS2024 main conference ended with such a racist remark by a faculty when talking about ethi….
0
268
0
Our results show our model can successfully capture the multi-modal transition model and is better suited for diverse and intricate motion synthesis tasks.
0
0
0
Another distinctive feature of AAMDM is its operation in a learned lower-dimensional latent space rather than the traditional full pose space, further accelerating the generation process.
0
0
0