Advanced-Project
Setup-Guide
Install treep and clone all required repos using treep
python3 -m pip install -U treep
git clone https://github.com/machines-in-motion/treep_machines_in_motion
treep --clone NYU_FINGEROnce you have the workspace folder you can build it using the following
source /opt/openrobots/setup.bash
source /opt/ros/foxy/setup.bash
colcon build --cmake-args -DCMAKE_BUILD_TYPE=Release -DPYBIND11_TEST=OFF -DBUILD_TESTING=OFF
source ~/Desktop/advanced/workspace/install/setup.bash
colcon build --cmake-args -DCMAKE_BUILD_TYPE=Release -DPYBIND11_TEST=OFF -DBUILD_TESTING=OFFDifferent versions
mirror_bot/local (deprecated)
Uses gravity compensation in only the leader bot and PD controller in the follower. Both bots needs to be connected to the same host. Also has the HW direct controller which directly sets the position and velocity using the Hardware, intead of calculating the tau.
mirror_bot/localv2
Uses gravity compensation in both the bots and and extra PD controller in the follower. Both bots needs to be connected to the same host. Using gravity compensation in the follower bot resulted in more precise control, it was not apparent at first due to the error being small, but the error can be measured accross controllers by logging the diff in leader and follower position over time.
mirror_bot/remote
Its same as the localv2 but can be used accross network and bots can be connected to different hosts. Uses sockets for communication. The performance drop from local to local network is negligible, further testing required for different network speeds and packet loss.
mirror_bot/replay
Just a simple script to record motion on any bot and replay the same motion using the log file.
mirror_bot/feedback
Force feedback from the follower bot so that the controller can feel if something is blocking the path of the follower. One issue currently is that there is a spring action in the leader which leads to a positive feedback loop if the leader is tensioned and then released. Not sure how can this be solved, since there are no force sensors to work with.
In all the leader uses, a P controller in addition to the gravity compensation we had in previous experiments. The follower is same as previous, a PD controller + gravity compensation.
Packet loss and delay simulation
We simulate packet loss using the linuc inbuilt netem tool. Since we are using local sockets, we can just use this tool on the loopback interface lo as follows:
tc qdisc change dev lo root netem delay 15ms 5ms 25% loss 1% 25%
The above means the following:
- 15ms +- 5ms delay with the next random element depending 25% on the last one. This isn't true statistical correlation, but an approximation.
- 1% of packets to be lost, and each successive probability depends by a quarter on the last one.