MeccaBLOG

Francesco Biral recently prepared a new poster for the SAFERIDER project, to be used for an interview with a local TV broadcasting network.

Here it is (click on the picture for the full poster).

Fabrizio Zendri is a Ph.D. student. Here is some excerpt from his current work.

RUMBy (Reduced-scale UnManned BuggY) is an internal project of the Mechatronics Lab aimed at the development of an autonomous vehicle able to drive in dynamic conditions (i.e. high speed). Hardware used in the project is extensively described in this pages; this post discusses the control system architecture (software). In RUMBy a Model Predictive Control (MPC) approach for controlling an autonomous vehicle is used.

The simplified schmatic architecture in Figure 1 describes the main elements of the autonomous vehicle guidance system and it is composed of three main modules: the trajectory/mode planning, the low-level control system , and the vehicle and environmental model.

The trajectory/mode planning module computes trajectory and mode of operation online during the drive, on the basis of current measurements, at fixed points or on the occurrence of certain events (such as tracking errors exceeding certain thresholds, presence of obstacles, etc..). It computes long parts of trajectory using a simplified vehicle model, a 5 degrees-of freedom (DOF) model also known as “bicycle model” and represented in Figure 2.

The low-level control system computes the optimal inputs (front steering angle, four brakes, engine torque) based on sensor measurements, states and parameters estimations and reference trajectory coming from the trajectory/mode planning module. The low-level control system objective is to keep the vehicle as close as possible to the currently planned trajectory despite sensor measurement noise, unmodeled dynamics, parametric uncertainties and sudden phenomena which are not taken into account by the trajectory planner. In RUMBy this module is based on an 8 DOF rolling vehicle model (Figure 3), which considers a 3D vehicle model with rolling DOF. In this system, parameters and states estimators are especially critical. The first one is based on recursive least square method and estimates quickly varying parameters, such as tire-ground contact coefficients or wind. The second one, in this case an Extended Kalman Filter (EKF), estimates vehicle interesting states (such as positions, yaw, velocities, forces, accelerationsetc…). These algorithms represent one of the most critical aspects in the project, because the controllability of the system depends on the accuracy the states are estimated with.

The third module, finally, can be of two different types: in simulation/prototyping phase, it is represented by a 14 DOF vehicle model that captures a lot of phenomena typical of real vehicles; in experimental phase, instead, it is the testing vehicle RUMBy. Actually some parts of the described architecture are available and others are “work in progress”. In particular, I’m actually developing a version of state estimator base on two kind of algorithms: a data preprocessing and an Extended Kalman Filter. The first one computes some operations to keep data as useful as possible: e.g. eliminating systematical effects (bias), assigning quality weights to measures (GPS data are useless with a reduced number of satellites), etc… The second one uses the corrected data to estimate vehicle positions, attitude and velocities.

SAFERIDER is a FP7 project for safety of motorcycles. The University of Trento participates with the role of developing ADAS (Advanced Driver Assistance Systems). Here is the official Fact Sheet .

Why did Rumby miss his goal? It was a little too close for his reduced maneuvering abilities. Here is what happens if the goal is twice as far (200 meters instead of 100).

Here are the effects of some tuning. Now “Frankestein” looks to have better control of his moves. The picture (click to enlarge) shows that the real trajectory departs very little from plans, except when the vehicle is close to his goal. The actual speed is also shown and compared to planned speed. On the speed profile the sequence of new plans is easier to see (recall that every while the “brain” updates motion plans, which allows to recover mismatches between planned and real motion). Motion ends when either the vehicle enters the circle surrounding the goal position, or the speed becomes zero (the latter is what happens in this case).

My friend Alberto Broggi (University of Parma) has been invited in Korea for a keynote speech with title: “VisLab’s footprint in the history of intelligent vehicles”. My compliments, Alberto!

(click to enlarge)

Here is one of the fist simulations of the closed loop behaviour of RUMBy. The system is made of a “body” (now still simulated, but it will be the real car model) and of a “brain”. The “body” is the car model, which sensors and low level data fusion and control loops. The brain is a computer, connected wireless to the body, which makes high level decision and especially the path planning with optimal control.

The body and the brain are here simulated and we are closing the loop. It looks like Frankestein awakening. The picture shows how Frankestein executes a navigation task. It has to start from the flag position, with initial velocity pointing upwards, and it has to reach the stop sign with velocity pointing right. Every while, the brain makes a motion plan, which is shown with the gray lines. The brain asks the body to execute that plan but the execution is imperfect, because the body does not react as the brain thought (there are mismatches between the model used by the brain for planning and the real dynamics of the body) and because there are time lags (the brain takes a while tio compute the plan and feeds it to the body with some lag). We can see, for example that the first plan asks to steer right more than the body really does. The red line is the actual executed trajectory. As the real motion drifts from plans, the brain is not discouraged, and makes continuously new plans to try to recover the situation. But “Frankestein” is somewhat drunk, and, as it approaches the goal, the execution becomes more and more critical. Eventually, Frankestein misses his goal.

But do not worry… that was a couple of days ago. We have now tuned the adrenaline and Frankestein behave a lot better. Stay tuned.

Here is the motorcycle driving simulator of the University of Padova. University of Padova (the Motorcycle Dynamic Research Group) is a partner of SafeRider Project. Their simulator will be used to test Advanced Driving Assistance Systems.

The Kick Off Meeting of SAFERIDER project began today in Thessaloniki, GR. SAFERIDER is a Framework Programme 7 research project aimed at active safety for motorcycles. A number of informative and active support systems (like those being developed for cars and trucks) have to be studied and developed for motorbikes. The 21 partners of this 36 months – 3.5 M€ projects are OEM (Original Equipment Manufactures) Yamaha and Piaggio, research Institutes and Universities and a number of suppliers (Ibeo, Metasystem, Avmap, NZI) and the Federation of European Motorcyclists’ Associations (FEMA).

Here is a movie of the Driving Simulator of the Bavarian Police School (one partner of Train All Project). The driving simulator is used to train police drivers. One important aspect is to manage the sometimes unexpected and unpredictable behaviour of some road users when facing emergency vehicles. The driving Simulator has a control room (shown in the end) which is used by traineers to instruct the trainee.