We introduce RAVEN, a 3D open-set memory-based behavior tree framework for aerial outdoor semantic navigation. RAVEN not only navigates reliably toward detected targets, but also performs long-range semantic reasoning and LVLM-guided informed search

⛔️Stop throwing away far range semantics, encode them as Rays instead ! 🔥Excited to present RayFronts at #IROS2025 in Hangzhou, China ! 🎥Catch us in the live presentation next Tuesday 16:45-16:50 Track 9.

This work was a collaborative effort with @OmarAlama, Dmytro Kurdydyk, John Keller, @Nik__V__ , Wenshan Wang, @ybisk , @smash0190 at @AirLabCMU

For more details, check out our paper on arXiv ( https://t.co/zXXvu42Ep1) and our project website: ( https://t.co/2XY4bEFxb7)

We further validated our method on a physical aerial robot equipped with a custom Orin AGX payload and ZED X camera, demonstrating real-world feasibility

We evaluated our method across 10 simulated outdoor environments and 100 semantic tasks, consistently outperforming state-of-the-art semantic navigation baselines.

Thanks to the task-agnostic design of our open-set semantic voxel-ray memory, RAVEN supports both multi-class query search and on-the-fly task switching.

When direct visual evidence is limited, we leverage LVLM-guided search to generate auxiliary object targets that are semantically related to the primary goals. These targets are seamlessly integrated into our ray-based search behaviors within the behavior tree

One of our key contributions is the use of "semantic rays", a direction-based representation that complements 3D voxels by encoding semantic information of far-range objects beyond depth coverage. This ray-based search enables the robot to plan effectively over long distances

Our behavior tree adapts robot's strategies: it performs semantic voxel search when reliable cues exist within short-range, switches to semantic ray search when only long-range directional hints are available, and invokes LVLM to suggest auxiliary cues to overcome sparsity

We address the challenges of outdoor semantic navigation: much larger spatial extent of outdoor scenes require long-range search, and sparse distribution of targets combined with absence of structured hierarchy demands careful strategic reasoning.

Meet MapAnything – a transformer that directly regresses factored metric 3D scene geometry (from images, calibration, poses, or depth) in an end-to-end way. No pipelines, no extra stages. Just 3D geometry & cameras, straight from any type of input, delivering new state-of-the-art

Want to learn how to empower 🤖 with real-time scene understanding and exploration capabilities? Catch Me, @hocherie1 & @QiuYuhengQiu presenting RayFronts at #RSS2025 SemRob Workshop (OHE 122) & Epstein Plaza at 10:00 am PST Today! https://t.co/yE90CQVU4y

Introducing UFM, a Unified Flow & Matching model, which for the first time shows that the unification of optical flow and image matching tasks is mutually beneficial and achieves SOTA. Check out UFM’s matching in action below! 👇 🌐 Website: https://t.co/0Cw9QyrefK 🧵👇

"Generalization means being able to solve problems that the system hasn't been prepared for." Our latest work in #RSS2025 can automatically invent neural networks as state abstractions, which help robots generalize. Check it out here: https://t.co/RkoR5MRRJg

🔥Best Paper Award at #ICRA2025 Thrilled to share that our paper MAC-VO has been awarded the 𝘽𝙚𝙨𝙩 𝘾𝙤𝙣𝙛𝙚𝙧𝙚𝙣𝙘𝙚 𝙋𝙖𝙥𝙚𝙧 𝘼𝙬𝙖𝙧𝙙 and the 𝘽𝙚𝙨𝙩 𝙋𝙖𝙥𝙚𝙧 𝘼𝙬𝙖𝙧𝙙 𝙤𝙣 𝙍𝙤𝙗𝙤𝙩 𝙋𝙚𝙧𝙘𝙚𝙥𝙩𝙞𝙤𝙣! Check our project: https://t.co/RvOT2aOmP9

RayFronts code has been released ! https://t.co/wecp43Gx8l 🤖 Guide your robot with semantics within & beyond depth. 🖼️ Stop using slow SAM crops + CLIP pipelines. RayFronts gets dense language aligned features in one forward pass. 🚀 Test your mapping ideas in our pipeline !

🚀 Thrilled to present ViSafe, a vision-only airborne collision avoidance system that achieved drone-to-drone avoidance at 144 km/h. In an era of congested airspace and growing autonomy, reliable self-separation is paramount 🧵👇

SIGLIP wins over CLIP even in dense tasks like zero shot open-vocab semantic segmentation on Replica . Using the RayFronts encoder (NA attention + RADIO @PavloMolchanov + SIGLIP @giffmana) projection to the CLS token gives you SoTA performance. No more SAM+CROP+CLIP business.

Excited to share RayFronts, an open-set semantic mapping system! 🚀 (1) Encodes semantics into “rays” at frontiers—great for outdoor nav, no fixed-depth limit. ⚡ (2) Fast dense vision-language encoder (tested on Orin AGX), achieves SOTA open-vocab 3D segmentation.