Marco Pavone
@drmapavone
Followers
4K
Following
71
Media
10
Statuses
191
Prof @Stanford, Distinguished Research Scientist and AV research lead @nvidia. PhD from @MITAeroAstro. Robotics, autonomous systems, AI. Opinions are my own.
Stanford, CA USA
Joined November 2018
RT @NVIDIADRIVE: .@drmapavone (@NVIDIA, @Stanford) joins the latest NVIDIA AI Podcast to break down how AI & simulation are critical for #A….
0
6
0
To learn more about Halos:. Web: Video: Blog: AV Safety Day: Podcast:
0
0
1
At #GTC2025, Jensen unveiled Halos, a comprehensive safety system for AVs and Physical AI. Halos integrates numerous technologies developed by my team @nvidia, and I was thrilled to help coordinate its launch alongside Riccardo Mariani and many amazing colleagues @NVIDIADRIVE.
1
12
28
Check out our latest work on #robotics safety, where we propose a CBF-derived safety filter, which can handle hundreds of simultaneous constraints while retaining real-time control rates. Great work by @danielpmorton @StanfordEng.
Introducing one of the fastest and safest robot controllers, for operational space and hierarchical tasks. Deploy your learned policies, or teleoperate your robot confidently with OSCBF. Website: Preprint: With @drmapavone
0
3
28
For the first time ever, @nvidia is hosting an AV Safety Day at GTC - a multi-session workshop on AV safety. We will share our latest work on safe AV platforms, run-time monitoring, safety data flywheels, and more! #AutonomousVehicles #AI at #GTC25 . ➡️
0
14
32
It was a pleasure and a honor to instruct the @usairforce Test Pilot School on the foundations of robot autonomy and physical AI: @StanfordEng.
0
1
10
Just published in @Nature_NPJ: we present a data-driven methodology for the high-performance control of continuum robots, including "squishy" robots :). Paper: Great collaboration with @ETH_en @GeorgeHallerETH.
Data-driven, nonlinear, SSM-based model-predictive control of soft robots outperforms other model reduction approaches by a large margin, as shown in Collaborators at @stanford and @ETH_en: @heateralora, @mattiacenedese, and @drmapavone
0
4
19
AI4I, the Italian Institute of Artificial Intelligence for Industry (, has launched an international call for Heads of R&D Units (. This is a unique opportunity to shape the AI roadmap in Italy and beyond! @FabioPammolli.
1
4
13
(5/n) Great project led by @JiaweiYang118 while he was an intern at @NVIDIAAI, in collaboration with Jiahui Huang, @iamborisi, @Yuxiao_Chen_, Yan Wang, @Boyiliee, @YurongYou, Apoorva Sharma, @MaxiIgl, @KarkusPeter, @danfei_xu, and @yuewang314.
0
0
7
(4/n) Together, DreamDrive and STORM make it possible to train in closed-loop AV / robotics stacks in general and highly realistic simulation settings -- a key step towards general robot autonomy.
1
0
4
(3/n) STORM leverages a data-driven Transformer architecture that directly infers dynamic 3D scene representations--parameterized by 3D Gaussians and their velocities--in a single forward pass.
1
0
3
(2/n) STORM solves these with a data-driven Transformer-based feed-forward model that:.✅ Reconstructs scenes from sparse data. ✅ Removes per-scene optimization and achieves 200ms reconstruction speed per scene (or 0.6s for a 20s video).✅ Learns data priors for generalization.
1
0
5
(1/n) Existing reconstruction methods struggle with:.1️⃣ Reliance on dense spatio-temporal observations. 2️⃣ Lengthy optimization times (hours). 3️⃣ Poor generalization.
1
0
4
Complementing DreamDrive, I am thrilled to introduce STORM, which enables fast scene reconstruction with a single feed-forward model. STORM transforms camera logs into dynamic 3D models - in real time!. Web: Paper:
1
31
126
4/n Great project led by @PointsCoder while he was an intern at @NVIDIAAI , in collaboration with @Boyiliee @iamborisi @Yuxiao_Chen_ Yan Wang, @YurongYou @ChaoweiX @danfei_xu and @yuewang314.
0
0
1
3/n As such, DreamDrive enables scalable driving scenario generation, and supports training and testing of driving models with in-the-wild data from anywhere in the world.
1
0
1
2/n DreamDrive combines video diffusion models with 3D Gaussian splatting to elevate visual references into dynamic 4D scenes. A hybrid Gaussian representation ensures consistent modeling of static and dynamic elements, creating realistic, trajectory-conditioned driving videos.
1
0
1
1/n DreamDrive tackles the challenge of synthesizing 3D-consistent visual observations from a single image. It enables scalable 4D scene generation directly from in-the-wild data without relying on costly data collection and annotations.
1
0
0
Introducing DreamDrive, which combines the complementary strengths of generative AI (video diffusion) and neural reconstruction (Gaussian splatting) to transform any street-view image into a dynamic 4D driving scene!. Web: Paper:
4
44
217
The @Stanford Aero/Astro Department is now inviting applications for a tenure-track faculty position. To learn more about the Department: To apply: @StanfordEng.
0
5
14