UniSim-SLAM: feed-forward SLAM with unified Sim(3) optimization lands at ECCV 2026
rsasaki0109 · x · 2026-09-07
Researchers at UNIST (Korea) present UniSim-SLAM (ECCV 2026), a feed-forward SLAM pipeline built on geometric foundation models that runs on phone-captured image sequences:
- Problem: feed-forward predictions depend heavily on the input view set, causing geometric inconsistency and trajectory drift over long sequences; online deployment trades two-view latency against multi-view constraint richness.
- Approach: lightweight two-view keyframe tracking in the frontend plus periodic multi-view submap refinement in the backend, combined in a unified multi-level factor graph on Sim(3) that jointly optimizes global keyframe poses and submap poses despite heterogeneous local scales.
- Key idea: treat two-view and multi-view predictions as complementary constraints—temporal two-view edges preserve tracking connectivity while multi-view submaps provide richer geometric anchors for long-term consistency.
Code is coming soon; a phone-capture demo is available.
More from Embodied
- Musk: Cybercab will run as 'some combination of Airbnb and Uber' fleet — Polymarket · 2026-09-07
- DIY outdoor robot U-BOT goes wireless on day 31, with a Claude-coded control app — _Stocko_ · 2026-09-07
- Uber logged 400,181 US misconduct reports in 5 years; Tesla touts driverless Robotaxi as the fix — SumitGup · 2026-09-07
- Rigging a 248-million-year-old extinct creature: a generalized 3D gen + skeleton pipeline in 6 steps — teortaxesTex · 2026-09-07
- A dog darted 15 feet in front of a Waymo in SF; it stopped instantly — julianweisser · 2026-09-07
- He Walked Away From a $7M Robot Dog Project: As AI Gets Stronger, the Scarce Skill Is Deciding, Not Prompting — yongqianme · 2026-09-07