WayTrace: A System Using Off-the-Shelf VLM, Onboard Trace Model, and RL Policy for Real-World Navigation
The ETH Robotic Systems Lab introduces WayTrace, a robotics system that converts web-map directions into real-world navigation tasks. The system utilizes an off-the-shelf Vision-Language Model (VLM), an onboard trace model, and a Reinforcement Learning policy to guide a robot through complex environments in Zurich, covering a total distance of 2.35 km. This demonstration includes navigating crosswalks, stairs, crowded streets, and construction sites.
The ETH Robotic Systems Lab presents WayTrace, a system combining an off-the-shelf Vision-Language Model (VLM), an onboard trace model, and a Reinforcement Learning policy to navigate 2.35 km in Zurich following web-map directions; the team attributes the demonstration to their own work while noting that specific technical details are available in the accompanying thread.
Source: ETH Robotic Systems Lab — X · Read original article ↗
Article text · Original source · English
Our new work WayTrace: an off-the-shelf VLM, an onboard trace model, and an RL policy take a robot from web-map directions to 2.35 km of real-world missions in Zurich.
Details in the thread 👇
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
distance traveled
2.35
Unique elapsed hours
View original evidence
Our new work WayTrace: an off-the-shelf VLM, an onboard trace model, and an RL policy take a robot from web-map directions to 2.35 km of real-world missions in Zurich.
Open source S2distance traveled
2.35
View original evidence
Then we let it find its own way through Zurich: crosswalks, stairs, crowded streets, construction sites, 2.35 km in total.
Open source S5
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
Our new work WayTrace: an off-the-shelf VLM, an onboard trace model, and an RL policy take a robot from web-map directions to 2.35 km of real-world missions in Zurich.
Open source S2
Then we let it find its own way through Zurich: crosswalks, stairs, crowded streets, construction sites, 2.35 km in total.
Open source S5
We gave a robot the walking directions you'd get from a web map. Then we let it find its own way through Zurich: crosswalks, stairs, crowded streets, construction sites, 2.35 km in total. Introducing WayTrace 🧵👇View the quoted post on X
Source：Robotic Systems Lab · x.com