PNP

This project is called 'PNP' for 'pick and place'. It can simulate a factory floor, where parts are picked up and placed down during various stages of a manufacturing process.

---

Table of Contents

• Video
• How to Run
• What
• Why
• How
• Next Steps

---

How To Run

Execute:

roslaunch brl_pxh_api brl_pxh_api.launch

Then in a separate window, execute:

roslaunch pnp pnp.launch

Video

Watch the project in action!

What

PNP can simulate any factory floor that can be described within the constraints of its JSON config file (/config/config.json).

For instance, PNP ships with an example floor that has two tasks: washing that takes 3 seconds, and welding that takes 2 seconds.

"tasks": [
    {
        "id": "wash",
        "duration": 3
    },
    {
        "id": "weld",
        "duration": 2
    }
]

Tasks are executed at stations. The same task can be executed at more than one station. For example, we see below that washing happens at both station_0 and station_2. This means that if a part needs to be washed, but station_0 is occupied, it can be handled in station_2.

"stations": [
    {
        "id": "station_0",
        "fid_id": 1,
        "task_id": "wash",
        "free": true 
    },
    {
        "id": "station_1",
        "fid_id": 2,
        "task_id": "weld",
        "free": true 
    },
    {
        "id": "station_2",
        "fid_id": 3,
        "task_id": "wash",
        "free": true 
    }
]

Thus, the mapping from tasks to stations can be one-to-many. Stations are represented by fiducial markers. We see above, for instance, that station_2 is paired with a fiducial of id 3. The physical fiducial is marked in red below.

A cargo represents the set of parts that are moved from one station to another during a manufacturing process. For example, a set of raw_parts is represented by cargo_0 below:

{
    "id": "cargo_0",
    "fid_id": 10,
    "type": "raw_parts",
    "cur_task_idx": -1,
    "last_station": "None",
    "last_placed": "Never"
}

And we see that a cargo of type raw_parts needs to be washed, welded, then washed:

"cargoTypes": [
    {
        "id": "raw_parts",
        "tasks": ["wash", "weld", "wash"]     
    },
    {
        "id": "washed_parts",
        "tasks": ["weld", "wash"]
    }
]

Physically, a cargo is represented in PNP by a 3D-printed cube fixed with a fiducial.

To see how these tasks, stations, and cargoes come together in a simulated factory, watch this recording of a demo.

Why

This project was developed by the Brandeis Robotics Lab as an example project for future students. The aim was to demo how one can use the lab's PX-100 robotic arm to implement an autonomous pick and place system.

How

The core of the vision and pick-and-place systems used by PNP were developed in an earlier project. Its README explains both.

What is new in PNP is:

1. the automation of the pick-and-place action; and
2. the scheduling algorithm that manages when and where to pick and place cargoes.

The first is achieved by communication between the pnp_frame_broadcaster node and the pnp_controller node. Whenever the former detects a cargo via the vision system, it publishes a message to the latter over the seen_cargoes topic. Next, the pnp_controller node picks up the detected cargo so long as it satisfies the constraints of its scheduling algorithm.

Also new in PNP is its use of the Brandeis Robotics Lab's API for the PX-100 arm rather than Interbotix's official API.

Next Steps

PNP can be readily extended. For example, we might add a feature that allows us to add a success rate to each task. Then, the challenge for PNP would be to automatically account for cases where a task fails.