real-time run

[celikemir]

Hi, thank you for the great work. Is it possible to run this algorithm on NVIDIA Jetson devices? I noticed an interesting slowdown mechanism where the algorithm runs at a slower rate during rosbag playback.

[tb78623]

Hii i am also facing the same issue on a Jetson device. Even i am able to run all the other algo's which are realtively more compute intensive the Adaptive Lio.

[chengwei0427]

Hi, @celikemir @tb78623
In adaptive_lio, the continuous time CT mode is used, involving multiple iterations and derivations of the initial and final poses, which is resource-intensive.

[tb78623]

@chengwei0427 Even after switching the ICP formulation to a point-to-plane model, I am observing no noticeable change in performance. Could you clarify which computational resources CT-ICP primarily relies on, and whether any of these are limited or not readily available on Jetson platforms? Additionally, what architectural or algorithmic modifications would you recommend to make CT-ICP run efficiently on a Jetson-class embedded system?”

[chengwei0427]

Hi， @tb78623
Adaptive-lio is resource-intensive in terms of data association, pose optimization, and map management. It can be said that adaptive-lio is merely an academic attempt; for practical use, I suggest you first debug the open-source ct-lio, as adaptive-lio is developed based on ct-lio.