A bicycle robot racing a mountain-bike trail in a Zurich forest. Roots, rocks, tabletop drops, all on onboard perception.
@rai_inst and ETH's @leggedrobotics. It raced 946 of the 969 m on its own, 2.3x faster than a person driving it by remote.
A model that flags a fall or an obstacle hit before it happens. Critical for outdoor robotics.
λ-Reachability, from @RuiChen_Rob and Changliu Liu's lab at CMU. Presented at #CoRL2026 this week.
🌱 How do you control a robot whose body is constantly growing, buckling, and reshaping?
📷 Put 19 cameras on it.
Meet PanoVine, the first autonomous vine robot system.
We distribute 19 cameras along a 6 meters, 7-DoF soft growing vine robot, giving it whole-body visual feedback of both itself and its surroundings. Using these observations, we train an end-to-end visuomotor policy that autonomously navigates branches, slopes, gaps, turns, and obstacles.
🔗 https://t.co/znQUj88VFu
Logging roads in Alberta, mapped from scratch, with a load of raw timber on the trailer.
@KodiakRobotics took the Kodiak Driver from highways and oil fields into timber country.
A Yamaha ATV at 10 m/s over loose gravel and dirt. The model looks at the terrain it's about to drive over.
SpatialBEV, from @micahnye31 at CMU AirLab. At #CoRL2026 this week.
A Coco sidewalk robot on just a monocular camera and coarse GPS, holding the walkway around obstacles and past pedestrians, day and night.
FlowPilot, from Honglin He and @zhoubolei at UCLA, hardware from @CocoRobotics. At #CoRL2026 this week.
A 6-meter soft vine robot with 19 cameras on its body, steering itself through branches, up a slope and across a gap.
PanoVine, from @SongShuran and Allison Okamura's labs at Stanford. @XiaomengXu11 presented it at #CoRL2026 in Austin this week.
Danny’s work at @reservoirfarms is something everyone in tech should be following.
Agriculture is a huge market, critical to human wellbeing and has been largely forgotten by San Francisco technologists.
Farmers are a part of the American backbone and they deserve more help from us.
Reservoir Farms run an incredible testbed down in the Central Valley trialling the coolest new agricultural technologies.
It’s always so cool to see what they’re testing on the farm.
From driverless grain carts in Iowa to a 20-ton cabless UGV in a Norwegian forest.
@Colin_Hurd started Smart Ag (AutoCart). Now his @MachAutonomy stack is hauling 5 tons on UGV Defence's FLEX L12 through mud and conifers. Ag autonomy travels well. 🌲
Real mine traffic. 60 driverless electric haul trucks working a copper mine at 5,400+ meters.
This @EaconMining clip from Zhibula hits hard: passing bays, a 3-way intersection, trucks taking turns at the dump.
Selective harvesting, 4,000 meters under the Pacific.
Impossible Metals' Eureka hovers over the seabed and picks nodules one at a time, skipping any with visible life on them. Ag robotics people will recognize the idea. 🌊
A robot arm in a rain poncho, a pallet of solar panels, and a crew in the mud.
@GrittRobotics is training on live job sites instead of a tidy lab. My kind of physical AI.
https://t.co/3cLi0hWOOd
Real dirt, frontier AI.
Most robotics research stays trapped in tidy benchmark labs. At Gritt, our models train on live, messy construction sites, feeding simulations that push our systems back into the field smarter every time.
Our Cofounder and CTO Vishal breaks down the real-to-sim-to-real flywheel building construction superintelligence 👇
@GrittRobotics
Everyone films the spraying. I like that @bonsairobotics showed the refill run.
OMC AR-500 working citrus, then driving itself back to the tank when it runs low.
Outdoor robotics is incredible.
You know a machine is getting interesting when this many people will stand in a bare field to watch it.
High Degree's steam machine at @reservoirfarms.