Odyssey straps 25-pound camera rigs to hikers to train world models

Most generative video startups are busy scraping YouTube and stock footage libraries. Odyssey is taking a far sweatier path: strapping 25-pound camera rigs to human hikers and sending them into the wild.

The startup announced an $18 million Series A funding round led by EQT Ventures, with participation from GV and Air Street Capital. The fresh capital brings Odyssey's total funding to $27 million, which it plans to spend scaling up its physical data collection across California before expanding further.

A 25-pound backpack full of LiDAR

Odyssey was founded by self-driving pioneers Oliver Cameron, formerly the VP of product at Cruise, and Jeff Hawke. Given their autonomous vehicle roots, it makes sense that they are obsessed with ground-truth sensor data rather than just pixels.

Their capture rig looks a lot like Google's classic Street View Trekker backpack. Weighing around 25 pounds, it packs six cameras, two LiDAR sensors, and an inertial measurement unit.

The idea is to send people anywhere a vehicle cannot go, like caves, hiking trails, beaches, glaciers, and dense forests. The hardware captures 360-degree footage at 3.5K resolution while tagging every frame with physics-accurate depth metadata.

From there, Odyssey feeds the stream into its algorithms to generate photorealistic digital reconstructions of the real world, somewhat similar to Meta's Hyperscape experiments.

Splitting reality into controllable layers

Capturing raw 3D scans is one thing, but turning them into promptable generative tools is another.

Cameron and Hawke say they are building models that deconstruct environments into distinct visual layers, including object geometry, environmental lighting, and dynamic motion. Creators can then manipulate each element independently to build customized virtual worlds.

Odyssey argues that training models on flat 2D internet video will never yield living, Hollywood-grade virtual environments. The company believes physical world models require rich, multimodal 3D data as foundational ground truth.

There are clear hurdles ahead. Public data capture always draws regulatory scrutiny over pedestrian privacy, a battle Google Street View fought for years. Odyssey will have to prove it can capture the world without running into legal walls.

Why it matters

We think Odyssey represents a fascinating divergence in how AI companies approach world models.

While OpenAI and others rely on internet-scale video to teach AI physics, Odyssey is applying the autonomous vehicle playbook. By gathering hardware-verified LiDAR and spatial depth data, they are building a proprietary data moat that web scrapers simply cannot touch.

For filmmakers, game developers, and VFX studios, the promise here is massive. Building high-fidelity 3D environments today takes months of manual digital sculpting and lighting passes. If Odyssey delivers a generative system that offers precise cinematic control over true 3D spatial assets, virtual production costs could plummet.

The real question is execution. Bridging the gap between static photogrammetry and fully controllable generative simulation is an unsolved technical challenge. Odyssey's physical-first approach is bold and refreshingly grounded in real hardware, but turning those scenic hikes into an intuitive creator platform remains an uphill climb.