ZJU-FAST-Lab / ego-planner-swarm

An efficient single/multi-agent trajectory planner for multicopters.
GNU General Public License v3.0
1.13k stars 222 forks source link

issue with path planning #27

Closed shubham-shahh closed 2 years ago

shubham-shahh commented 2 years ago

Hi, @bigsuperZZZX I am having an issue with tall and narrow obstacles, The planner tries to search path from top and bottom rather than finding paths from sides (favourable for this case). How can solve this

all your help is appreciated

thanks

bigsuperZZZX commented 2 years ago

Could you paste a picture here about what your obstacles look like?

shubham-shahh commented 2 years ago

Hi @bigsuperZZZX , here's a picture of the obstacle and the occupancy grid Screenshot from 2022-03-05 18-40-53

bigsuperZZZX commented 2 years ago

That's because the obstacle sensor only senses a limited region. From the mapped obstacle in the right figure, the produced trajectory is reasonable as it chooses the shortest path to fly through because the drone takes the unknown part as free.

shubham-shahh commented 2 years ago

That's because the obstacle sensor only senses a limited region. From the mapped obstacle in the right figure, the produced trajectory is reasonable as it chooses the shortest path to fly through because the drone takes the unknown part as free.

@bigsuperZZZX , That's correct, but, it will keep moving upwards as it maps the obstacle, but it is not good for such kind of obstacles, what are the possible solutions, to tackle such obstacles. I have mounted a lidar with 90 degree vertical FOV for obstacle detection

thanks

bigsuperZZZX commented 2 years ago

It's a complicated problem to tackle obstacles outside the sensing region. There are different candidate strategies, such as predicting obstacle shape using neural networks (paper), adjusting trajectory to see more known places (paper), or just using wide range sensors like fisheye cameras (paper), etc.

shubham-shahh commented 2 years ago

It's a complicated problem to tackle obstacles outside the sensing region. There are different candidate strategies, such as predicting obstacle shape using neural networks (paper), adjusting trajectory to see more known places (paper), or just using wide range sensors like fisheye cameras (paper), etc.

Hi, Thanks for the response. I'll look into them