New demo, with turns and swerves, of distributed real-time multi-agent planning/MPC; no central control needed. Uses GBP and p2p message passing over the joint factor graph so arbitrarily scalable and robust. Dyson Robotics Lab arxiv.org/abs/2203.11618
Conversation
Replying to
This is the same method with slightly different settings: all agents have forward goals, and only forward/back acceleration is allowed.
Quote Tweet
Using Gaussian Belief Propagation as in Robot Web, we now show dynamic multi-robot *planning* via p2p comms, no central solver needed. In a scaled simulation, cars slide closely past each other at motorway speeds. @AalokPat @rmurai0610, Dyson Robotics Lab, arxiv.org/abs/2203.11618
Show this thread
0:24
11.7K views
13
Does this simulate any latency and sensor noise or is it all simulation perfect and instant ?
1
At the moment this assumes all robots have perfect state knowledge of each other; the only task is to plan future movement steps. See however our related Robot Web work on distributed state estimation with noisy sensor measurements arxiv.org/abs/2202.03314.
1
1
4
Show replies
The planning is based on the assumption that you can accurately detect the trajectory of the other agents. That might not be the case. The planning should be jointly optimized with the multi-object tracking.
1
4
Yes, agreed. We have also worked on distributed multi-robot localisation (see Robot Web), but we haven't yet put localisation and planning together
1
2
Show replies
Looks like a virtual version of Shibuya crossing in Japan 😆
GIF
1
7
Or pretty much any African or Asian country.
GIF
read image description
ALT
Hello Andrew, this is looking pretty impressive. Every time I see a GBP demo it makes me want to learn more.
They're on the left side of the road. English traffic 😂
1
