Autonomous Navigation in Urban Environments
Authors
Calvet I Molinas, Joan ; Kooverjee, Himal ; Roca, David Romanos
Term
4. term
Education
Publication year
2018
Submitted on
2018-06-07
Pages
150
Abstract
På Aalborg Universitet forestiller vi os en selvkørende transportservice på bestilling (Autonomous Mobility on Demand, AMoD). Dette speciale lægger grundlaget ved at udvikle og afprøve de grundlæggende byggesten på en lille, tilgængelig platform: en golfvogn. Først laves en fysikbaseret model af vognen til realistisk simulering samt en forenklet model, der gør reguleringsdesign håndterbart. For at knytte modellerne til den virkelige platform modelleres også aktuatorerne (styring, gas og bremsning). Med dette designes ikke-lineære regulatorer, der får vognen til at følge en given bane, og de valideres i simulation; derefter tilpasses de med aktuator-modellerne og køres på den faktiske golfvogn. Da regulatorerne kræver præcis tilbagemelding, samles målinger fra de ombordværende sensorer med et Udvidet Kalman-filter (EKF) for at estimere vognens position og bevægelse. Den kombinerede regulator og EKF testes på platformen. Til planlægning af ruter gennemgås metoder på forskningsfronten, og en optimal variant af Rapidly-exploring Random Trees (RRT*) med kinodynamiske begrænsninger implementeres og evalueres i simulation. Planlægningsmiljøet opbygges som et optagethedskort (occupancy map) – et gitter, der markerer fri og besat plads – konstrueret ud fra to LIDAR-laserscannere ved hjælp af Cartographer-biblioteket i Robot Operating System (ROS).
At Aalborg University, we imagine an on-demand self-driving transport service (Autonomous Mobility on Demand, AMoD). This thesis lays the groundwork by developing and testing the basic building blocks on a small, accessible platform: a golf cart. We first create a physics-based model of the cart for realistic simulation, and a simplified model to make control design tractable. To connect models to the real vehicle, we also model the actuators (steering, throttle, and braking). Using these, we design nonlinear controllers that keep the cart on a given path and validate them in simulation, then adapt them with the actuator models and run them on the real cart. Because the controllers require accurate feedback, measurements from the onboard sensors are fused with an Extended Kalman Filter (EKF) to estimate the cart’s position and motion. The combined controller and EKF are tested on the platform. For path planning, we survey state-of-the-art methods and implement an optimal variant of Rapidly-exploring Random Trees (RRT*) with kinodynamic constraints, evaluating it in simulation. The planning environment is built as an occupancy map—a grid that marks free and occupied space—constructed from two LIDAR sensors using the Cartographer library within the Robot Operating System (ROS).
[This abstract was generated with the help of AI]
Keywords
Nonlinear control ; Autonomous ; Navigation ; Urban ; Environments ; Path ; planning ; planner ; sensor fusion ; EKF ; Kalman ; RRT ; Lyupanov redesign ; Backstepping ; Vehicle ; golf cart ; MoD ; Mobility-on-Demand
Documents