Tractor_bot

[mfocchi]

https://answers.ros.org/question/249874/tracked-robot-gazebo-ros/

https://pdfs.semanticscholar.org/9dbc/830fc9b8e85b8c84a2d993d40fa947e005ca.pdf

https://github.com/progtologist/gazebo-tracks

[mfocchi]

• [x] change base_link frame in coppeliasim to match with urdf
• [x] change right_front_wheel frame in coppeliasim to match with urdf
• [x] meshlab : simplify all meshes
• [x] timed coppeliasim simulation with ROS
• [x] do experiments with motion capture system downstairs (alpha does not matter)
• [x] implement torque control (talk with company)
• [x] identification for low speed to eval w_act without having the artifact of high slippage when a wheel slows down
• [x] test outdoor to identify slip = k + alpha exp(-betaR)
• [x] check why you had divergence in last experiments (if Vdot remains negative )
• [ ] RRT * with dubins with minimum time cost function
• [x] (Davide) Approccio 2: valuta curva pareto
• [x] (Michele) Approccio 1: optimal control
• [x] (Michele) add filters to computation of alphadot
• [x] (davide) implement code for test on real robot
• [x] Test with more challenging trajectories
• [x] init with dubins curves the optimization
• [x] update side slip model in optimation
• [x] solve algebraic loop and evaluate sensitivity
• [ ] add offsets on alpha and compensate them
• [ ] simulink test adding noise on position estimate (optitrack)
• [x] (Michele) publish Ref, actual, alpha, alphadot, v, omega, v_bar, omega_bar
• [ ] To create a more accurate simulator replace coppeliasim contact computation with the force obtained from shear displacement
• [x] (Michele) set frequency to 200Hz and repeat the experiments with all logs
• [x] (Michele) automatize estimation cycle and do it before any batch of experiments
• [x] (michele) create structure for the paper
• [x] (michele) maxxi_interface publish at 11Hz instead than 200Hz
• [x] (Michele) Fix optitrack x y sign theta
• [ ] (michele) Trajectory generation that slows down when error increases
• [ ] (Nutu/Mohamed) outdoor identification tests. Identify slippage parameters starting from ident test, validate the model with different terrains
• [x] (Nutu) create service call in ros2 to call the optimization code to generate a slippage aware trajectory
• [ ] (michele) simulink test with parameter uncertainty
• [ ] Implement adaptive approach where you estimate slippage and compensate for it
• [ ] Implement robust invariance set evaluation
• [ ] compare P0) planning with dubins P1) optimal control (v const) planning with unicycle P2) optimal control planning with full model C1) control unicycle C2) control full model and do the combination of these
• [x] (Nutu/Mohammed) implement slippage compensation in C++ and perform single tests at v = 0.1 and different omega to see the effect of slippage compensation
• [ ] (Nutu/Mohammed) test connecting via ethetnet robot to optitrack router and do rosbag to check delays
• [x] (Nutu) design and 3d print an elevation fixture for markers
• [x] (Nutu) Improve the transfer of optimized trajectory from matlab to C++
• [x] (Mohammed) put Z1 arm on tractor
• [x] (Mohammad) do identification tests in simulation
• [x] (Michele) implement alphadot=0 controller in locosim
• [ ] -fix ros2-maxxi add velocity to jointstate and debug function getSpeedrads
• [x] (rocco galati) fix delay and restore baud rate
• [ ] (Nutu/Michele) gravity compensation
• [x] (Michele) add planner
• [ ] - [ ] (Michele) compare with zou
• [ ] Input state stability analysis
• [ ] (Michele) delay robust controller
• [x] (Nutu) Implement gravity compensation. check in coppelia sim if it is enough to compensate gravity with the validated model
• [x] (luigi) c1 c2 parameter estimation with Lyapunov
• [ ] (Nutu) Experiment outdoor with Santoro
• [x] (Michele) Heightfield 30deg
• [ ] (michele) input state stability analysis
• [x] (Nutu) implement slam with zed2 without CUDA
• [ ] (michele) Enrico Mingo whole body control
• [x] (angelo/mohammad) repeat ID experiments indoor varying wheel speed instead than omega v
• [x] fixed baud rate
• [ ] testing ouside with gravity compensation and slam on slopes
• [x] (riccardo) Kalman filter for slippage
• [ ] (michele) implement RK 4 in tracrtorsim
• [ ] extend model to slopes and to have torque inputs instead of velocity
• [x] reidentify alpha with new simulator
• [x] try different traj with matlab in sim showing is exploiting friction if you set it very high and use the model with slippage to optimize
• [x] NN with accel input for beta

[mfocchi]

PAPER

• [x] SIM add initial state different from the reference $x(0) = 0.05$, $y_d = 0.02$, $\theta_d=0.1$.
• [x] redo Fig 1 and 2 add xb yb axes
• [x] add downstairs identification tests
• [ ] repeat identification of alpha for a meaningful turning radius
• [ ] do sim with a velocity profile that is inline with the ID and one not showing the latter is worse
• [x] add figure with matlab sim showing with alpha dot it converges to zero
• [x] add noise to inputs
• [ ] do video of the robot.

FUTURE WORKS

• test with gravity compensation on slopes (we found out that kinematic model is enough to see the effect (on flat) we see in reality), extend model to 3D,

• identify fr and add rolling friction

• now we can 1) estimate the slippage and make the robot behave as a unicycle and use dubins for rrt * and 2) create better traj considering slippage with optimal control

• slippage map learning with a NN

• 3D slippage map navigation

[mfocchi]

pf = [1.5; 1.0; -1.2];%pf = [10.0; 10.0; -1.9]; % works with LONGSLIP

left Unicycle right LONGSLIP model

SIDE LONG SLIP MODEL

[mfocchi]

SIMULATION: no long slippage no side slippage control:

no long slippage, side slippage control:

long slippage, side slippage: