Parking project

Advanced driver assistance systems (ADAS) are one of the fastest-growing segments in automotive electronics. In particular, according to analysts , autonomous valet parking systems will play an important role in next years. The following are the main benefits for such systems:

  •   They enhance the comfort of the driver, reducing stress related to finding a free place and performing a, possibly complex, parking maneuver. They also decrease the overall travel time.
  •   They increase safety, preventing the vehicle from collisions during the maneuver, with fixed obstacles, other vehicles or pedestrians.
  •   They allow a better usage of available space in parking areas.
  •   They reduce the fuel needed for the parking operation.

     
    Currently, the solutions commercially available for automatic parking are not satisfactory, due to the following limitations:

  •   They allow only assisted parking; the driver must be present in the vehicle during the operation.
  •   They are not flexible: the vehicle must be properly aligned at the beginning of the maneuver, in a position close to the final parking position; otherwise, the maneuver may not be possible.
  •   The search of a free parking slot is left to the driver.
  •   They are not completely automatic; the control on brake and acceleration is left to the driver.

     
    The main advantages of the proposed solutions are the following ones:

  •   The parking is completely autonomous and does not require the presence of the driver.
  •   The vehicle is able to autonomously find an available parking place.
  •   The parking planner is able to accept any initial configuration from which there is a feasible maneuver to the desired parking slot, even if this involves complex trajectory planning.
  •   The vehicle is able to take active obstacle avoidance measures.
  •   The proposed device is cheap to add to a passenger car, since expensive multiple-layer laser scanners are avoided in favor of cheaper stereo vision systems.
  •   The parking operation is fast, since the maneuver is based on a minimum-time solution. This allows increasing the efficiency and the overall throughput of the automated parking lot.

     
    European automotive industry is demanding now a major improvement on existing technologies. The increasing speed of processors and the diminishing costs of hardware allow the embedded implementation of advanced sensing, planning and control methods.
     
    Addressing these issues is a very challenging task:

  •   It requires efficient and precise sensing and data fusion from sensors that produce large data.
  •   It requires efficiently finding the minimum-time solution of a motion-planning problem for a nonlinear model (the vehicle), with state and input constraints (the obstacles and the maximum allowed values for velocity, acceleration and steering).
  •   It requires the design of a robust tracking controller, based mainly on visual information.

     
    AURORA Lab. has the competences to address this problem. In particular, it has a strong experience in:

  •   path planning for nonholonomic vehicles,
  •   path following for nonholonomic vehicles and mechanical systems,
  •   control of underactuated mechanical systems with the virtual constraint approach,
  •   minimum-time control with input and output constraints.
  •   solution of nonlinear, nonconvex optimization problems.

     
    The project will have a significant impact for European car industry and will be a step towards the development of system for automated highway driving. In fact, sensing, data fusion and trajectory planning are crucial tasks for fully automated drive, that according to various roadmaps, will be developed by 2025.The project aims at developing an experiment prototype at an advanced technological readiness level. The experiment will consist of a fully functional prototype, operating in an environment that reproduces a typical urban parking scenario

The design will be based on recent results obtained by UNIPR in trajectory generation, minimum-time constrained control, state tracking and nonlinear optimization together with state of the art methods for perception, image processing and sensor fusion. More specifically:

  •   The motion planning problem (that amounts at finding a trajectory that joins the initial and fi- nal configuration, avoiding the obstacles and respecting the maximum curvature requirement due to robot kinematics) will be solved using methods based on the finite-element solution of the Hamilton-Jacobi-Bellman equation.
  •   The resulting path will be smoothed by interpolation with eta-3 splines to obtain a high degree of geometric continuity for a smooth actuation.
  •   The speed law along the found path will be determined as the solution of a minimum-time constrained problem.
  •   The feedback control policy will be implemented by iterative replanning, using Kalman filtering techniques for reconstructing the state from the sensor signals (odometry, ultrasound sensors and vision). The information given by the various sensors will be integrated to obtain precise localization information.

     
    Under a scientific point of view, the results of the project will be relevant since they will show that it is possible to solve numerically the HJB equation to obtain a real-time solution of a motion-planning problem for a constrained system with nonholonomic constraints. It is important to note that this method provides a deterministic solution to the optimal planning problem, differently from randomized methods, such as rapidly exploring random tree (RRT), more commonly used in the solution of motion planning problems.