Kristijan Maček

A reinforcement learning approach to obstacle avoidance of mobile robots

One of the basic issues in the navigation of autonomous mobile robots is the obstacle avoidance task that is commonly achieved using a reactive control paradigm where a local mapping from perceived states to actions is acquired. A control strategy with learning capabilities in an unknown environment can be obtained using reinforcement learning where the learning agent is given only sparse reward information. This credit assignment problem includes both temporal and structural aspects. While the temporal credit assignment problem is solved using core elements of the reinforcement learning agent, solution of the structural credit assignment problem requires an appropriate internal state space representation of the environment. In this paper, a discrete coding of the input space using a neural network structure is presented as opposed to the commonly used continuous internal representation. This enables a faster and more efficient convergence of the reinforcement learning process.

Motion Planning for Car-Like Vehicles in Dynamic Urban Scenarios

This paper focuses on development of a motion planning strategy for car-like vehicles in dynamic urban-like scenarios. The strategy can be summarized as a search for a collision-free trajectory among linearly moving obstacles applying rapidly-exploring random trees (RRT) and B-splines. Collision avoidance is based on geometric search in transformed state space of chained form kinematic model decomposition. The time criterion for avoiding obstacles is based on relative robot to obstacle motion and is checked iteratively for possible collisions within the RRT exploration phase. The line segment geometric path is interpolated with a B-spline curve in order to generate a feasible trajectory that takes into account nonholonomic constraints. The exploration strategy aims at finding an optimal steering and longitudinal control of the vehicle in minimum time and steering activity sense. In order to test the strategy a MatLab based simulator was developed. This simulator reproduces a simple 2D urban-like environment with parked and moving cars, buses, trucks, people, buildings, streets, and trees. The test vehicle, a modified smart car equipped with several sensors was kinematically modeled. The sensor data are extracted from the environment based on its geometrical description and used as input data for the motion planning strategy which was verified in a dynamic urban scenario simulation

Dynamics modeling and parameter identification for autonomous vehicle navigation

This paper focuses on development of a dynamic model for an Ackermann-like vehicle based on a static tire-road friction model and laws of technical mechanics. The model takes as input the steering angle of the wheels in front and the rotational velocities of the drive wheels in the back of the vehicle. It delivers a 3-DOF output in terms of CoG vehicle velocity, body slip angle and the yaw rate of the vehicle in the x-y plane, as well as estimates on the forces acting on the system. It is suitable for modeling dynamic vehicle regimes in e.g. overtaking maneuvers/obstacle avoidance and lane-keeping, enabling active steering control by stabilizing the dynamics of the vehicle. The physical model description is based on previous works combined with a suitable friction model that is tractable in practice. Experimental verification of the obtained model is given for the Smart testing vehicle platform, where a separate analysis is done for directly measured as opposed to estimated/optimized parameters of the model.

Safe Vehicle Navigation in Dynamic Urban Scenarios

This paper has presented the deliberative part of the navigation architecture for the SmartTer platform, comprising two main component: (a) route planning, which finds a set of configurations between two given points of the environment while taking into account given traffic rules; and (b) partial motion planning, which handles the actual execution of the plan while taking into account the dynamic elements of the world. A key aspect of the navigation architecture proposed is that a special attention is paid to the motion safety issue, ie the ability to avoid collisions. Different safety levels are explored and their operational conditions are explicitly spelled out (something which is usually not done). The results depict safe navigation in dynamic scenarios whichinclude both moving vehicles and pedestrians, where the architecture has been implemented in both real and simulated platforms, although only simulation results are provided at the time of paper writing. On the theoretical side, an interesting research direction is the exploration of more advanced motion prediction techniques in order to improve both the accuracy and the time horizon of our safety checks.

Path following for autonomous vehicle navigation with inherent safety and dynamics margin

This paper addresses the path following problem for autonomous Ackermann-like vehicle navigation. A control strategy that takes into account both kinodynamic and configuration space constraints of the vehicle, denoted as traversability-anchored dynamic path following (TADPF) controller is presented. It ensures secure vehicle commands in presence of obstacles, based on traversability information given by a global navigation function. By additionally using a reference point on the global smooth path, the local vicinity path configuration with respect to the vehicle is taken explicitly into account to ensure smooth and stable path following. Furthermore, a previously developed sliding mode path following (SMPF) controller that results in fast convergence rate and low path following error but which does not consider kinodynamic constraints, is augmented by the the kinodynamic and configuration space constraints check of the TADPF controller. The new proposed control strategy denoted as TADPF-SMPF controller thus combines advantageous characteristics of both original control strategies for path following, yielding inherent safety and vehicle dynamics margin. All three control strategies are verified in simulation, whereas the TADPF and TADPF-SMPF path following schemes are also verified experimentally.

An approach to motion planning of indoor mobile robots

In this paper we present a motion planning approach of indoor mobile robots based on integration of A* path planning algorithm and dynamic window local obstacle avoidance method. A simple and efficient procedure to the selection of appropriate motion commands based upon alignment of trajectories generated by the dynamic window module and the global geometric path is proposed. Global occupancy grid map is incrementally updated in on-line manner. The algorithm is verified in Saphira simulated environment for differential drive Pioneer 2DX mobile robot (manufacturer ActivMedia Robotics) using laser range sensor.

2D laser-based probabilistic motion tracking in urban-like environments

All over the world traffic injuries and fatality rates are increasing every year. The combination of negligent and imprudent drivers, adverse road and weather conditions produces tragic results with dramatic loss of life. In this scenario, the use of mobile robotics technology onboard vehicles could reduce casualties. Obstacle motion tracking is an essential ability for car-like mobile robots. However, this task is not trivial in urban environments where a great quantity and variety of obstacles may induce the vehicle to take erroneous decisions. Unfortunately, obstacles close to its sensors frequently cause blind zones behind them where other obstacles could be hidden. In this situation, the robot may lose vital information about these obstructed obstacles that can provoke collisions. In order to overcome this problem, an obstacle motion tracking module based only on 2D laser scan data was developed. Its main parts consist of obstacle detection, obstacle classification, and obstacle tracking algorithms. A motion detection module using scan matching was developed aiming to improve the data quality for navigation purposes; a probabilistic grid representation of the environment was also implemented. The research was initially conducted using a MatLab simulator that reproduces a simple 2D urban-like environment. Then the algorithms were validated using data samplings in real urban environments. On average, the results proved the usefulness of considering obstacle paths and velocities while navigating at reasonable computational costs. This, undoubtedly, will allow future controllers to obtain a better performance in highly dynamic environments.

MOTION PLANNING IN THE PRESENCE OF MOVING OBSTACLES USING RRT SEARCH AND B-SPLINES

Reference LSA-CONF-2006-020 Conference Web Site: http://www.syroco2006.deis.unibo.it/index.html Record created on 2006-12-07, modified on 2016-08-08

Discussion on: “Adaptive and Predictive Path Tracking Control for Off-road Mobile Robots”

The authors are addressing the problem of accuratepath tracking in off-road terrains in presence oflateral sliding, with particular application to agri-culture vehicles. The adherence conditions are notmodeled explicitly (i.e. no tire to ground model),however the lateral sliding angles of the wheels areintegrated in the kinematic bicycle model, calledextended kinematic model. An observer is used toestimate the unknown sliding angles on-line withrespect to steady-state adherence conditions. A steer-ing wheel controller based on chained form trans-formation of the kinematic bicycle model for theparametric curve tracking is used. Furthermore, inorder to compensate for abrupt curvature changesand low level control delay, a model predictive con-trol signal is added. Experimental verification isprovided on slippery, wet ground for straight driv-ing on a slope and curved path driving on flat ground,with good tracking results achieved, close to setrequirements.

SMART Navigation in Structured and Unstructured Environments

Recently, intelligent transportation systems have been introduced for tasks like automated parking and highway driving. This is one of many contact points between human and robot intelligence, in that a human driver is sharing the driving task with intelligent computer systems. In this video we present an automated passenger vehicle that is able to autonomously navigate through both structured and unstructured B23environments without relying on prior environmental information or known waypoints. The system uses ego motion estimation based on an inertial measurement unit and internal vehicle sensors, and combines this with a laser range finder to map its environment. It uses a combination of global planning and local planning to safely navigate through the environment to a desired goal location.