Briefing

Sample Based Motion Planning. Work in Progress.

This project is intended as educational replication of several general ideas:

• Sparcity
• Motion Primitives
• Importance Sampling

A sub-goal of this project is to integrate and leverage benefits of several of these ideas in a hybrid solution.

Inputs to program

• system dynamics and various constraints are supplied as functions
• environment obstacles

Progress:

• Sparcity:
  • implemented core algorithm of Stable Sparse RRT (https://www.cs.rutgers.edu/~kb572/pubs/stable_sparse_rrt_WAFR14_LLB.pdf)
  • approximate nearest neighbour with stochastic search, optional compile flag for linear nearest neighbour search
• Motion Primitives:
  • lookup for feasible control for steering toward a direction (https://arxiv.org/pdf/1809.02399.pdf)
  • compile flag for enabling its use;
  • adapted for run-time lookup filling
  • non-grid based greedy goal-neighbourhood search
  • [TODO] sparse lookup storage
• Importance Sampling:
  • [TODO] use of entropy based sampler for shifting towards better parameterization (https://journals.sagepub.com/doi/pdf/10.1177/0278364912444543)

Running Planner

• prerequisites
  • install Rust: https://rustup.rs/
  • internet connection (building project pulls in software dependencies)
  • Make (to build triangulation tool if using custom maps, or obtain executable at: https://www.cs.cmu.edu/~quake/triangle.html)
• build and run in release mode
  • Either:
    • have custom maps already generated (see Generating Custom Maps section)
    • -p <problem instance name> (eg: -p obs3 ), see prob_instances.rs for predefined problem domain list
  • Or:
    • -o <file_obstacle>: obstacle file path (eg: -o obstacles/obs2.txt (randomly generated boxes)
  • Or:
    • have custom maps already generated (see Generating Custom Maps section)
    • -e <.ele file path> -n <.node file path> (see custom maps section) -m <model>
• optional arguments:
  • -w: show witness node and witness representative pairs
    • drawn as a line(red) with end points (purple: witness), (blue: witness representative)
  • -i <N>: max iterations
  • -m <model>: model selection, defaults to dubins (some parameters overriden by src/prob_instances.rs)
    • <model> variants:
      • dubins, airplane (TODO)
  • -b <N>: batch N iterations in between rendering calls
  • -h: help
• optional compile-time features:
  • usage:
    • cargo run --release --bin planner --features nn_naive,disable_pruning,(other features...) -- -p <problem_instance_name> (other program arguments)...
  • variants:
    • motion_primitives (enables motion_primitive)
    • runge_kutta (alternative RK4 propagation method, default is Euler)
    • disable_pruning (make the propagation tree non-sparse)
    • nn_naive (use linear search for nearest neighbour query)
    • nn_sample_log (use logarithmic number of stochastic samples for nn query, must not have nn_naive to have effect, defaults to square root number of stochastic samples)
    • mo_prim_debug (render all candidates for motion primitives)
    • mo_prim_thresh_low/high (low and high neighbourhood threshold for motion primitive activation)

Generating Random Obstacles (a couple obstacles exists in obstacles/ folder)

• build and run in release mode with: cargo run --release --bin gen_obs -- -f <output_file_path>
• required arguments:
  • -f <output_file_path> (eg: cargo run --release --bin gen_obs -- -f obstacles/obs99.txt)
• optional arguments:
  • -n <N>: number of obstacles to be generated (default: 30)

Using Random Obstacles

• cargo run --release --bin planner -- -o <file_obstacle>
  • -o <file_obstacle>: obstacle file path (eg: -o obstacles/obs2.txt)

Generating Custom Maps (need to run this once in order to use custom maps):

• a set of maps that is mainly used for benchmarking purposes obtainable from https://www.movingai.com/benchmarks/grids.html can be used, these are located in the /maps_custom folder
• character movable space within a map are triangulated for use in the planner as the configuration free space
• triangulation is done using the awesome Triangle software from http://www.cs.cmu.edu/~quake/triangle.html
• the maps are converted into a format for Triangle to process and output is loadable into our planner and further extruded as triangular prisms for use with a general purpose 3D obstacle detector, these intermediate files are stored at /maps_custom//poly
• some maps might have bad triangulation not useable for the planner (I aimed for working with Dragon Age maps)
• generating intermediate files and map assets for our planner
  • 1st, compile Triangle
    • cd Triangle_v1_6
    • make
  • 2nd, run ./script_map2poly.sh (generates formatted file for Triangle, may take a while)
  • 3rd, run ./script_triangulate_poly.sh (outputs 2D triangulation result as .ele and .node files)
  • all set for use...

Using Custom Maps

• build and run planner with custom maps in release mode:
  • Either (recommended): cargo run --release --bin planner -- -p <problem_instance_name> (see src/prob_instances.rs)
  • Or: cargo run --release --bin planner -- -e <.ele file path> -n <.node file path> -m <model>(edit appropriate model file for different source/goal states)
    • (eg: cargo run --release --bin planner -- -e maps_custom/dragon_age/poly/ost100d.1.ele -n maps_custom/dragon_age/poly/ost100d.1.node -p ost100d -i 500000 -m dubins

Screenshots

Randomly Generated Boxes

Maps from Dragon Age