Installation and Pre-requisites

Install CARLA

See here - https://carla.readthedocs.io/en/0.9.12/start_quickstart/

Important: The code is cimpatible with CARLA version 0.9.12

IS-DESPOT-p

1. A C++11 compatible compiler is required. IS-DESPOT-p has been developed using g++ 7.3.0. Newer version might work too.

   sudo apt-get install build-essential

2. Makefile

   sudo apt-get install make

3. Navigate to Carla-CTS02/ISDESPOT/isdespot-ped-pred/is-despot/ and run make

4. Navigate to Carla-CTS02/ISDESPOT/isdespot-ped-pred/is-despot/problems/isdespotp_car and run make. This will create a binary car.

Python environment

Run pip install -r requirements.txt

In order to replicate the results from our experiments, the train and test files need to be executed.

Before running the code make sure your working directory is Carla-CTS02

HyLEAR

• For training run python train_hylear.py --shared --cuda --port=2000

--shared parameter enables sharing of weights between actor and critic networks of the learner in HyLEAR, --cuda switches pyotrch to use cuda, port=200 defines the carla port

• For testing run python eval_hylear.py --shared --port=2000 --agent=hylear --test="12"

--port=2000 defines carla port, --test="12" is an optional parameter and can be used if testing on a single scenario, by default tetsing will be done on all scenarios.

NavSAC-p

• For training run python train_sac.py --shared --cuda --port=2000
• For testing run python eval_sac.py --shared --port=2000 --test="11"

--test="11" is an optional parameter required only if testing on a single scenario

A2C-CADRL-p

• For training run python train_a2c.py -p=2000

where -p=2000 defines the carla port

• For testing run python eval_a2c.py --test="01"

similar to testing for HyLEAR, --test is an optional parameter required only if testing on single scenario

IS-DESPOT-p

• Since no training is required for IS-DESPOT-p, for testing run python eval_isdespot.py --despot_port=1245 --test="01" --agent="isdespot"

--despot_port=1245 defines the port to communicate with the planner, --test is the same as HyLEAR and --agent=isdespot defines whether to use IS-DESPOT-p or IS-DESPOT-p*. In order to use IS-DESPOT-p*, pass the argument --agent=isdespot*

HyLEAP

• For training run python train_hyleap.py --despot_port=1250
• For testing run python eval_hyleap.py --despot_port=1250 --test="11"

--despot_port=1250 defines the port to communicate with the planner, --test is an optional parameter same as in HYLEAR

HyPAL

• For training run python train_hypal.py --shared --cuda --port=2000
• For testing run python eval.py --shared --port=2000 --agent=hypal --test="11"

License

Shield:

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.