Heng Yang
@hankyang94
Followers
5K
Following
755
Media
57
Statuses
311
Assistant Professor @Harvard SEAS @hseas, Lead the Harvard Computational Robotics Lab. #Robotics, #Optimization, #Control, #Vision, #Learning
Cambridge, MA
Joined January 2017
Sharing a project that’s kept me excited for months:. Five years ago, I tried projecting a 10000×10000 symmetric matrix onto the positive semidefinite cone using MATLAB’s eig on my MacBook—gave up out of sheer impatience. Today, we released a CUDA-based factorization-free method
6
35
330
Check out Kevin’s work on Adapting Conformal Prediction to Distribution Shifts without Labels at UAI!.
Excited to present our paper at #UAI2025 this Wednesday on adapting prediction sets (i.e. conformal prediction) to arbitrary distribution shifts with no need for new labels!
0
0
5
Excited about this work that leverages simulation for more efficient real-world validation.
Can we use simulation to validate Physical AI? Yes—with far fewer real-world tests. We propose a control variates–based estimation framework that pairs sim & real data to dramatically cut validation costs. #AI #Robotics #Sim2Real". Paper: @NVIDIADRIVE.
0
0
3
Wow "Building Rome with Convex Optimization" is a #RSS2025 Outstanding Systems Paper Award Finalist — congratulations to @Hyhan0118!. Also excited to see XM tackled a massive bundle adjustment problem involving 20,000 images, shared by a GitHub issue — an SDP of size 60,000 ×
"Building Rome with Convex Optimization" has been accepted to #RSS2025!. Try XM, our new structure from motion pipeline powered by GPU-accelerated convex semidefinite optimization:. XM solves large-scale (nonconvex) global bundle adjustment problem via
0
21
138
"Building Rome with Convex Optimization" has been accepted to #RSS2025!. Try XM, our new structure from motion pipeline powered by GPU-accelerated convex semidefinite optimization:. XM solves large-scale (nonconvex) global bundle adjustment problem via
1
44
292
RT @lucacarlone1: Here is the recording of the Robotics Worldwide Workshop, held at MIT on April 4, 2025: .-include….
0
26
0
Going to #ICLR2025? .Check out our Spotlight paper "Control-oriented Clustering of Visual Latent Representation" presented by Han Qi. TL;DR: We investigated the geometric structure of the visual latent space in a vision-based control pipeline trained via behavior cloning, and
0
7
49
Had fun presenting @ShuchengK's recent work, “Local Linear Convergence of ADMM for Solving SDPs under Strict Complementarity,” at the Numerical Analysis Seminar hosted by the Department of Mathematics at the University of Maryland. In this work, we uncover and prove an
0
2
16
In Contact-Rich motion planning, the so-called "Contact-Implicit" formulation stands out as it embeds discrete and combinatorial contact sequencing using implicit "complementarity constraints" in a continuous optimization problem. The contact-implicit formulation looks
0
10
70
RT @zhenjun_zhao: Building Rome with Convex Optimization. @Hyhan0118, @hankyang94. tl;dr: 2D keypoints+pretrained depth->scaled BA->QCQP->e….
0
11
0
Here is the paper containing technical details of . “Global Contact-Rich Planning with Sparsity-Rich Semidefinite Relaxations”. Stay tuned for SPOT (Sparse Polynomial Optimization Toolbox)!.
How robust can model predictive control be if we can solve each trajectory optimization to global optimality?. On the contact-rich push-T problem, we show that model-based global optimization is so robust that it never fails, even if the model is not even correct!. We achieve
1
10
55
Diffusion has shown great promise for generating robot **actions**, can it act as a **world model** to generate the future conditioned on actions?. In our work led by @hanqi359246 @hcy1n and in collaboration with @du_yilun, we show a **controllable** action-conditioned video
8
50
282
How robust can model predictive control be if we can solve each trajectory optimization to global optimality?. On the contact-rich push-T problem, we show that model-based global optimization is so robust that it never fails, even if the model is not even correct!. We achieve
6
34
179
Congrats to our wonderful Rhodes Scholar Aneesh!. Made us proud!.
2024 Wrapped🎁:.Won the Rhodes Scholarship (@rhodes_trust)!.Presented first-author work at #NeurIPS!.Named a Top 10 @harvardtech Innovator!.Co-led new Preprint: Robots!.Survived PhD apps!. None of this happens without my mentors; grateful beyond words!
0
0
19
How is Rosalind going to face the hundreds of talented, hardworking, and genuine Chinese students at @MIT when she got back?. They (and many of us) went through so much trouble to get to the US due to the pure drive to do good science, and yet you put this in your slide. This is.
Mitigating racial bias from LLMs is a lot easier than removing it from humans! . Can’t believe this happened at the best AI conference @NeurIPSConf . We have ethical reviews for authors, but missed it for invited speakers? 😡
17
14
317
RT @aneeshers: Just landed and here are at #NeurIPS2024!! Would love to chat with anyone on: lifelong RL (FastTRAC), generative world model….
0
3
0
Going to miss #NeurIPS2024 but please check out Aneesh's "Fast TRAC: A Parameter-Free Optimizer for Lifelong Reinforcement Learning"!.
⭐New Paper Alert ⭐.How can your #RL agent quickly adapt to new distribution shifts ? And without ANY tuning?🤔. We suggest you get on the Fast TRAC🏎️💨, our new Parameter-free Optimizer that surprisingly works. Why?. Website:1/🧵
0
1
20
Sparse Polynomial Optimization with Unbounded Sets just accepted to SIAM Journal on Optimization!. Congrats to @ShuchengK!. A lot to expect from sparse moment-sos relaxations!.
The sparse Moment-SOS hierarchy just got even more powerful -- handle unbounded sets!. With this, we solve optimal control of the unstable Van der Pol oscillator to certifiable global optimality!. Great work led by Lei, @ShuchengK, and Jie!.
0
1
23