
Faculty at @MIT. Learning for control and optimization. Reinforcement Learning. Multi-agent systems. Autonomy. Mobility.
Cambridge, MA
Joined April 2007
- Tweets445
- Following871
- Followers2.5K
- Likes291
This project was 4 years in the making and it's finally out!
We found that controlling vehicle speeds to mitigate traffic across a city can cut carbon emissions by 11-22%. To do this, we used deep reinforcement learning to optimize one million eco-driving scenarios. ๐๐ค๐ง
Cathy Wu (Hiring PhD/postdocs) retweeted
๐The era of autonomous multi-agent discovery is arriving! @karpathy
๐ชธExcited to share CORAL, our new work on autonomous multi-agent systems for open-ended scientific discovery.
๐
โโ๏ธA key limitation of many current โself-evolvingโ frameworks is that agents still operate inside tightly constrained loops โ they mutate solutions, but they do not truly decide how to explore.
In CORAL, we push toward genuine autonomy:
Agents decide
๐ what to explore
๐ง what knowledge to store
โป๏ธ which ideas to reuse
๐งช when to test hypotheses
๐ฅOne of the most interesting findings:
A single autonomous agent already outperforms fixed evolutionary search, but the biggest gains emerge when multiple agents form a research community.
๐ชOver 50% of breakthroughs in multi-agent runs come from building on other agentsโ discoveries. This suggests that knowledge reuse and collaboration are central to scalable automated discovery.
๐
Across 10+ difficult tasks in algorithmic discovery and system optimization, CORAL achieves state-of-the-art performance while improving efficiency by 3โ10ร.
๐ Paper: arxiv.org/abs/2604.01658v1
๐ป Code: github.com/Human-Agent-Socieโฆ
๐กAlphaXiv: alphaxiv.org/abs/2604.01658
#agentic #llms #selfevolvingagent #multiagent #autoresearch #alphaevolve
If pure AI or pure search isn't working for you, consider learning-guided optimization!
This approach combines the best of both worlds; our most recent work demonstrates this for coordinating agents in warehouse robotics.
Code & Video demo: github.com/MikeZheng777/RL-Rโฆ
MIT News: news.mit.edu/2026/ai-system-โฆ
Paper (JAIR): jair.org/index.php/jair/artiโฆ
This wouldn't have been possible without an amazing team. Huge shoutout to Han Zheng (lead author), Yining Ma, Brandon Araki, Jingkai Chen, and Symbotic!
#DeepRL #Robotics #WarehouseAutomation #AI
Have you noticed how navigation apps include walking & waiting for public transit, but excludes parking & walking for driving? After being late a few times ๐
, we finally did. We got curious: what if these apps account for parking?
Our resulting paper, Probability-Aware Parking Selection, shows that doing so can improve travel time estimates, save drivers time by routing them to parking, and make for fairer comparisons across transportation modes! And this work is featured today on the MIT Home Page!
More here: news.mit.edu/2026/parking-awโฆ.
Paper: arxiv.org/abs/2601.00521
Code: github.com/chickert/Probabilโฆ
Joint work with lead author Cameron Hickert, Sirui Li, Zhengbing He
#transportation #multimodal #parking #navigation #dynamicprogramming
๐ฃ ๐ฃ๐ผ๐๐๐ฑ๐ผ๐ฐ ๐ผ๐ฝ๐ฒ๐ป๐ถ๐ป๐ด๐ ๐ถ๐ป ๐บ๐ ๐ด๐ฟ๐ผ๐๐ฝ!
Iโm looking to hire ๐ญโ๐ฎ ๐ฝ๐ผ๐๐๐ฑ๐ผ๐ฐ๐ to work with us across the groupโs research areas, including:
๐ค Deep reinforcement learning
๐ฆ Traffic modeling & control
๐ง Neural combinatorial optimization
Join us in our mission to use ๐๐/๐ ๐ ๐๐ผ ๐๐ฎ๐ฐ๐ธ๐น๐ฒ ๐ต๐ฎ๐ฟ๐ฑ ๐ผ๐ฝ๐๐ถ๐บ๐ถ๐๐ฎ๐๐ถ๐ผ๐ป ๐ฝ๐ฟ๐ผ๐ฏ๐น๐ฒ๐บ๐ and to enable ๐ฏ๐ฒ๐๐๐ฒ๐ฟ, ๐ฒ๐๐ถ๐ฑ๐ฒ๐ป๐ฐ๐ฒ-๐ฑ๐ฟ๐ถ๐๐ฒ๐ป ๐ฑ๐ฒ๐ฐ๐ถ๐๐ถ๐ผ๐ป๐ ๐ถ๐ป ๐ฟ๐ฒ๐ฎ๐น-๐๐ผ๐ฟ๐น๐ฑ ๐๐ผ๐ฐ๐ถ๐ผ๐๐ฒ๐ฐ๐ต๐ป๐ถ๐ฐ๐ฎ๐น ๐๐๐๐๐ฒ๐บ๐.
๐๏ธ Applications are reviewed ๐ผ๐ป ๐ฎ ๐ฟ๐ผ๐น๐น๐ถ๐ป๐ด ๐ฏ๐ฎ๐๐ถ๐, starting immediately.
๐ Details + how to apply: cathywu.github.io/prospectivโฆ
Please share with anyone who might be a great fit!
๐ฃ Inviting all in the neural combinatorial optimization community to submit to our upcoming NeurIPS workshop!
๐ฅ New workshop at @NeurIPSConf!
DiffCoALG bridges the gap between classic algorithms & differentiable learning.
Think: LLM reasoning, routing, SAT, MIP โ neurally optimized end-to-end.
๐ Submit by Aug 22!
๐ sites.google.com/view/diffcoโฆ
#NeurIPS2025
Website: vindulamj.github.io/eco-drivโฆ
Published paper: sciencedirect.com/science/arโฆ
Pre-print: arxiv.org/abs/2408.05609
This work would not have been possible without lead author @vindulajay and collaborators Baptiste Freydt, Ao Qu, Cameron Hickert, Edgar Sanchez, Catherine Tang, Mark Taylor, Blaine Leonard, and financial support of UDOT, @NSF, and Amazon AWS, and support from MIT SuperCloud.
Are you a transportation practitioner or researcher who has read "Confessions of a Recovering Engineer" by Charles Marohn? And attending #TRBAM in a couple weeks?
I found this scathing critique of traffic engineering to be thought provoking, and would love to discuss it.
Cathy Wu (Hiring PhD/postdocs) retweeted
The Climate Change AI workshop at NeurIPS tomorrow will be livestreamed for free, even if you weren't able to register for NeurIPS (which was oversubscribed this year)!
Another highlight of our #NeurIPS2024 workshop on Sunday will be a panel discussion on small vs big models in ML for climate impact, featuring Chris Bretherton, Anamika Dubey & @BDilkina.
Stream the whole workshop live & for free via our website: climatechange.ai/events/neurโฆ
Want to learn more about neural combinatorial optimization (NCO)? Informal meetup in an hour at #NeurIPS2024, followed by lunch! 11am, West exhibition hall, near Door A. #ORMS Details: calendar.app.google/NEd7HMqmโฆ









