Self Drive Level 4, Skateboard

​

Preet Shah

Advisor: Bob Paulose Appackan

(Director and Tech Lead at Engineering Services International)

​

​

Abstract

• A skateboard chassis is basically a drive by wire chassis, with a drive system, and battery pack inbuilt.
• These were originally designed to aid rapid development of various body types on a single chassis development.
• Over the years, the advancements in manufacturing and smaller factories, facilitating larger customisation for each vehicle has generated a lot of interest in such chassis.

Similar Chassis, developed by Pixmoving

Software Stack for Skateboard

1) Path planning Algorithm 2) Mapping

Understanding the Environment

• Our environment comprises of a grid generated in Gazebo.
• The Skateboard has to navigate through the world, given the starting and ending points.
• In the path there might be obstacles which the skateboard has to avoid and find an alternative route to reach destination.
• The global planning will be done with the help of a GPS.

​

Tasks involved in Motion Planning

• Navigation is the problem of finding a collision-free motion for the robot.
• Coverage is the problem of passing a sensor or tool over all points in a space, such as in demining or painting.
• Localization is the problem of using a map to interpret sensor data to determine the configuration of the robot.
• Mapping is the problem of exploring and sensing an unknown environment

Exploring an Open Space

Global Path Planning Approach used

• There are many approaches used for global path planning suh dijkstra, rrt, a* formation etc;
• For our application, we used RRT* algorithm.
• It has an upper edge over other algorithms because of its speed and implementation.
• It tends to create more straight paths
• Useful especially in dense field of obstacles.

RRT* - Algorithm

• Sample random point
• Extend closest point towards sample, if no obstacles
• For all points in neighbourhood of extension, rewire such that the cost of traversal till extension is least.

RRT*-Obstacles

• Polygonal Intersections are computationally expensive
• Massive Speed Boost

Improvised RRT*

• Make the system flexible, and path find the joint system using RRT*
• 5 Dimensional configuration space
• 2 dimensional task space

RRT* Over a Long Distance between starting and Ending Points

Local Path Planning

• Explored some of the open source approaches like bug algorithms, visibility graph method, voronoi diagrams, cell decomposition method, tangent graph, path velocity decomposition, potential field approach, and TEB_local_planner.
• Potential field approach seemed to be one of the most efficient ways to avoid obstacles.

Drawbacks of Potential Field Approach

• Depends on chosen potential function. Finds the collision free path for chosen function, not in accordance with environment.
• Local Minima Problem - When attractive forces are balanced by repulsive forces, it becomes stationary.
• In a narrow corridor, it experiences repulsive forces simultaneously from opposite sides, and consequently the motion becomes unstable.

robot

Obstacles

It is stationary as forces are balanced

Teb_Local_Planner

• Teb implements an online optimal local trajectory planner for navigation and control of mobile robots.
• It is flexible in a way such that it is able to switch to current globally optimal trajectory among the candidate set.
• Formulation of sequence of n+1 robot poses linked together from an initial configuration to a goal:

• which can be solved by means of a least-squares non-linear solver. The terms wk are weights used to balance the different contributions, and ek,i the i-th cost contributions of objective k during the interval ∆ti .

Parallel Planning Enabled

I am speed :)

• Customising the optimization, we find that trajectory cannot transit across obstacles.
• Parallel planning specifically needs to get enabled for distinctive topologies

Optimizing multiple trajectories in distinctive topologies

• Gives current best trajectory, accounting for cheapest optimization cost.

Via Points

• Investigating the optimizer behaviour under the presence of via-points.
• Trajectory gets attracted by a via-point.
• But it does not reach the point.

• Shifting the waits for via points parameter.
• The problem occurred due to 2 conflicting objectives:
• minimum-time trajectory (time optimality)
• minimize the distance to the via-point

Giving it a tough time

​

Adding more via-points and obstacles closby to see the performance.