try base_local_planner instead of dwa_local_planner

[marc-hanheide]

No description provided.

[marc-hanheide]

then we need to test parameters and find a stable set

[Jailander]

We (me and @cdondrup ) just did some testing on this, the base_local_planner was not an improvement, the problem is that it rejected a lot of goals without even trying, and it sometimes quitted the navigation when an obstacle first appeared, even though overall the trajectories were smoother.

Seeing this we decided to try different parameter configurations on the dwa_planner, the first finding was the you can deactivate the dwa approach by setting the use_dwa parameter to false this led to a smother behaviour, but the robot was unable to go across doors anymore, so we decided to play with the parameters a bit and found that setting min_trans_vel: 0.0 made the robot turn on the stop and led to a generally more robust behaviour.

I think this needs further testing, do you think we can get anyone else to verify this. I leave the final parameter configuration for the dwa planner on our tests

DWAPlannerROS: {
  acc_lim_th: 2.0, 
  acc_lim_theta: 3.2,
  acc_lim_x: 1.0,
  acc_lim_y: 0.0,
  acc_limit_trans: 0.1,
  angular_sim_granularity: 0.1,
  forward_point_distance: 0.325,
  goal_distance_bias: 9.0,
  holonomic_robot: false,
  latch_xy_goal_tolerance: true,
  max_rot_vel: 1.0,
  max_scaling_factor: 0.2, 
  max_trans_vel: 0.55,
  max_vel_x: 0.55,
  max_vel_y: 0.0,
  min_rot_vel: 0.2,
  min_trans_vel: 0.0,
  min_vel_x: 0.0,
  min_vel_y: 0.0,
  occdist_scale: 0.01,
  oscillation_reset_angle: 0.2,
  oscillation_reset_dist: 0.05,
  path_distance_bias: 5.0,
  prune_plan: true,
  restore_defaults: false,
  rot_stopped_vel: 0.1,
  scaling_speed: 0.25,
  sim_granularity: 0.025,
  sim_time: 1.7,
  stop_time_buffer: 0.2,
  trans_stopped_vel: 0.0,
  use_dwa: true,
  vth_samples: 20,
  vx_samples: 3,
  vy_samples: 10,
  xy_goal_tolerance: 0.3,
  yaw_goal_tolerance: 0.1
}

[cdondrup]

Just a small addendum, we also set the trans_stopped_vel to 0.0 as a consequence of the above changes. Due to the turning on the spot, it also is able to plan out of corners that it usual gets stuck in.

[Jailander]

yes true thank you @cdondrup

[marc-hanheide]

Excellent, good job! Many, may thanks.
I think the new parameters should be tried by other sites now. @cburbridge @bfalacerda @nilsbore @hawesie @yianni @lucasb-eyer @ToMadoRe @mzillich you think you want to give this a try and put your observations in here, please? I leave this open until every site has reported their level of happiness and then we shall make the found parameters default in strands_movebase

[nilsbore]

On it! I have similar experiences with the base_local_planner. Didn't explore too much of the parameter space though.

[lucasb-eyer]

This sounds just right, esp. the *trans*vel part. I'll try to test it soon-ish, though I'm swamped right now.

[cburbridge]

I have not had a chance to test these parameters yet, but surely we should just have vy_samples: 1 to (presumably) only sample the current y velocity? This might also mask strands-project/strands_morse#96.

[lucasb-eyer]

ping @vonovak

[bfalacerda]

i got lost here, have these params been merged already?

[nilsbore]

I think they have, and it's working fine here.

[nilsbore]

Or.. Looking at https://github.com/strands-project/strands_movebase/blob/indigo-devel/strands_movebase/strands_movebase_params/dwa_planner_ros.yaml, it doesn't seem like it, in particular I don't see trans_stopped_vel and min_trans_vel.

[bfalacerda]

@Jailander @cdondrup could you make a PR so I can put it in bob more easily?

[Jailander]

Ok

[bfalacerda]

in the place we have bob nav works well anyway, so we can't really see the benefits of these params. Given https://github.com/strands-project/g4s_deployment/issues/8 , maybe we should at least try them in simulation? Can someone do it? @kunzel ?

[lucasb-eyer]

In our place they may be extremely beneficial, but I don't have the time to test them until in two and a half weeks, deadlines and meetings in the other project...

[vonovak]

The robot does behave better with the *trans*vel = 0

Also: Setting the forward_point_distance parameter to 0 improves how the robot is turning on the spot. Instead of oscillating left and right in some situations, it can make a full ~180 deg. turn. Also, increasing angular_sim_granularity to about 0.4 sometimes made the robot on-the-spot rotations a bit less jerky.