Junwei Liang 梁俊卫

@JunweiLiangCMU

Assistant Professor @HKUST (GZ) // Ph.D. @CarnegieMellon // NeurIPS / CoRL / ICLR / ICML Area Chair

Guangzhou, China

junweiliang.me

Joined September 2015

139 Following

841 Followers

Replies Reposts Media

Pinned
Junwei Liang 梁俊卫
@JunweiLiangCMU
Sep 6
Excited to share that our HKUST(GZ) lab has 3 papers accepted at CoRL 2026! 🚀 #1 Meet the Perceptive Behavior Foundation Model (Perceptive BFM). 🤖 Our paper adapts human motion priors to robot-centric terrain using onboard vision, all **without changing** the raw kinematic
5
Junwei Liang 梁俊卫
@JunweiLiangCMU
Oct 31
The teleoperation seems quite smooth and it is not easy to make the remote communication work!
Alexander Koch
@alexkoch_ai
Oct 29
You can now book our humanoid cleaners on a recurring subscription, for up to 3 hours per cleaning at tau-robotics.com. Video is first take, uncut, shown at 4× speed. See the reply for detailed footage at 1× speed.
Junwei Liang 梁俊卫
@JunweiLiangCMU
Oct 16
H2 SONIC is now open-source! 🤖 Our new work, TRACE, adapts the pretrained SONIC controller from G1 to the full-sized H2 in just ~8 hours on a single RTX PRO 6000 GPU. A new robot body without training from scratch! 💰 We combine joint and dynamics alignment with motion data
9
Junwei Liang 梁俊卫
@JunweiLiangCMU
Oct 15
ATHLETE is here! Teaching a humanoid to play tennis with low-cost motion data 🎾🤖 We used just one 10-minute tennis practice video for the striking motions, plus 16 walking sequences from a public dataset. Our trajectory-guided motion matching combines them into whole-body
2
Junwei Liang 梁俊卫
@JunweiLiangCMU
Sep 25
DiT4DiT is accepted at NeurIPS 2026! 🎉 🧠 Jointly trains Video + Action Diffusion Transformers to turn learned video dynamics into robot actions. ⚡ Extracts rich video features in a single forward pass for efficient closed-loop control. 🏆 98.6% success on LIBERO and 56.7% on
1

Junwei Liang 梁俊卫

Junwei Liang 梁俊卫’s posts