BTPG

KO | EN

gitlite — search

#python #java #python3 #arduino #golang #machine-learning #rust #html #flask #javascript #seismology #nodejs

★ 17 Open GitHub ↗

Behavior Tree Planning Gym

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# BTPG (IJCAI 2025)

Based on RoboWaiter, [VirtualHome](http://virtual-home.org/) v2.3.0 and OmniGinson.
<div align="center">

[[Appendix]](images/_IJCAI_25_BTPG_Appendix.pdf)

![Python Version](https://raw.githubusercontent.com/DIDS-EI/BTPG/main/images/python310.svg)
![GitHub license](https://raw.githubusercontent.com/DIDS-EI/BTPG/main/images/license.svg)

______________________________________________________________________

</div>
## 🛠️ Installation

### Create a conda environment

```shell
conda create --name btpg python=3.10
conda activate btpg
```

### Install BTPGym

```shell
cd BTPG
pip install -e .
```

ubuntu用户安装依赖
```
sudo apt-get update && sudo apt-get install -y libxcb-xinerama0 libxcb-icccm4 libxcb-image0 libxcb-keysyms1 libxcb-randr0 libxcb-render-util0 libxcb-xkb1 libxkbcommon-x11-0 graphviz
```

### 1. Download the VirtualHome executable for your platform (Only Windows is tested now):

| Operating System | Download Link                                                                      |
|:-----------------|:-----------------------------------------------------------------------------------|
| Linux            | [Download](http://virtual-home.org/release/simulator/v2.0/v2.3.0/linux_exec.zip)   |
| MacOS            | [Download](http://virtual-home.org/release/simulator/v2.0/v2.3.0/macos_exec.zip)   |
| Windows          | [Download](http://virtual-home.org/release/simulator/v2.0/v2.3.0/windows_exec.zip) |

### 2. Download the RoboWaiter executable (Only Windows is tested now):

👉 [Download RoboWaiter](https://drive.google.com/file/d/1ayAQZbPOyQV2W-V_ZdYv6AoqLOg0zvm1/view?usp=sharing)

Download the simulator, unzip it, and run `CafeSimulator.exe` to open the simulator. The simulator will display an empty scene, awaiting the code to generate the scene and complete robot interactions.

### 3. Download the OmniGibson executable (Only Windows is tested now):

We use the [isaacsim 4.2](https://docs.omniverse.nvidia.com/isaacsim/latest/index.html) simulator and [omnigibson](https://behavior.stanford.edu/omnigibson/getting_started/installation.html) dataset (omnigibson platform is not needed).

First download the isaacsim simulator and assets: 
| Operating System | Download Link                                                                      |
|:-----------------|:-----------------------------------------------------------------------------------|
| Linux            | [Download](https://download.isaacsim.omniverse.nvidia.com/isaac-sim-standalone%404.2.0-rc.18%2Brelease.16044.3b2ed111.gl.linux-x86_64.release.zip)   |
| Windows          | [Download](https://download.isaacsim.omniverse.nvidia.com/isaac-sim-standalone%404.2.0-rc.18%2Brelease.16044.3b2ed111.gl.windows-x86_64.release.zip) |

| Assets | Download Link                                                                      |
|:-----------------|:-----------------------------------------------------------------------------------|
| assets            | [Download](https://storage.googleapis.com/gibson_scenes/og_assets_1_1_0.tar.gz)   |
| og_dataset          | [Download](https://storage.googleapis.com/gibson_scenes/og_dataset_1_0_0.tar.gz) |

Put or link the isaacsim-4.2 and omnigibson_assets to the following path:

```shell
ln -s /path/to/isaacsim-4.2 /path/to/BTPG/simulators/omnigibson/IsaacSim4.2
ln -s /path/to/omnigibson_assets /path/to/BTPG/simulators/omnigibson/Assets
```

Install python packages in isaacsim 4.2:

```shell
cd BTPG
./simulators/omnigibson/IsaacSim4.2/kit/python/python3 -m pip install pyzmq lz4
```

## 📂 Directory Structure

```
btpg
│
├── agent - Configuration for intelligent agents.
├── algos - Training and decision-making algorithms.
├── bt_planning - Behavior tree planning algorithms.
│   ├── ReactivePlanning 
│   ├── BTExpansion
│   ├── OBTEA
│   └── HOBTEA
├── llm_client - Modules for large language model integration.
│   └── vector_database_env_goal.py - Core vector database functionality.
├── behavior_tree - Behavior tree engine components.
├── envs - Scene environments for agent interaction.
│   ├── base - Foundational elements for environments.
│   ├── gridworld - Grid-based testing environment.
│   ├── RoboWaiter - Café service robot scenario.
│   ├── VirtualHome - Household robot scenario.
│   ├── RobotHow - Testing environment for household robots.
│   └── RobotHow_Small - Smaller version of the household robot testing environment.
└── utils - Supporting functions and utilities.
```

```
simulators - Platforms for realistic training environments.
│
├── robothow - RobotHow simulator.
├── virtualhome - VirtualHome simulator.
└── omnigibson - IsaacSim4.2 and omnigibson dataset.
│   ├── IsaacSim4.2 - IsaacSim4.2 simulator.
│   ├── Assets - OmniGibson assets.
│   │   ├── assets - Unzip the [assets](https://storage.googleapis.com/gibson_scenes/og_assets_1_1_0.tar.gz) in here.
│   │   ├── og_dataset - Unzip the [og_dataset](https://storage.googleapis.com/gibson_scenes/og_dataset_1_0_0.tar.gz) in here.
│   │   └── omnigibson.key - This will be created automatically when you first use the omnigibson datasets.
```

```
test_exp - Testing modules for behavior trees planning, LLMs, and scene environments.
```

## 🚀 Usage

1. Download the simulator ([windows version](http://virtual-home.org/release/simulator/v2.0/v2.3.0/windows_exec.zip))
2. Unzip all files in windows_exec.v2.2.4 and move them to simulators/virtualhome/windows.
3. Run the test_exp/main.py and see the simulation result.
```python
python test_exp/main.py
```

## 📖 Getting Started
HOBTEA uses OpenAI's GPT-3.5 as the language model. You need to have an OpenAI API key to use HOBTEA. You can get one from [here](https://platform.openai.com/account/api-keys).

After the installation process, you can run BTPG by:

```python
import btpg
import time
from btpg import BehaviorTree
from btpg.utils.tools import *
from btpg.algos.bt_planning.main_interface import BTPlannerInterface
from btpg.algos.llm_client.tools import goal_transfer_str

# Initialize environment and conditions
scene = "VH"  # RW RH RHS
env, cur_cond_set = setup_environment(scene)
goal_str = 'IsIn_milk_fridge & IsClose_fridge'

# Transfer goal string to goal set
goal_set = goal_transfer_str(goal_str)  # [{'IsIn(milk, fridge)', 'IsClose(fridge)'}]

# Initialize Behavior Tree Planning Interface
algo = BTPlannerInterface(env.behavior_lib, cur_cond_set=cur_cond_set,
                      priority_act_ls=[], key_predicates=[],
                      key_objects=[],
                      selected_algorithm="hobtea", mode="big",
                      act_tree_verbose=False, time_limit=15,
                      heuristic_choice=0, output_just_best=True)
# Process goal set
algo.process(goal_set)
time_limit_exceeded = algo.algo.time_limit_exceeded
ptml_string, cost, expanded_num = algo.post_process()

# Execute Behavior Tree
error, state, act_num, current_cost, record_act_ls, ticks = algo.execute_bt(
    goal_set[0], cur_cond_set, verbose=False
)

# Output results
print(
    f"\x1b[32m Goal: {goal_str} \n Executed {act_num} action steps\x1b[0m",
    "\x1b[31mERROR\x1b[0m" if error else "",
    "\x1b[31mTIMEOUT\x1b[0m" if time_limit_exceeded else ""
)
print("Current cost:", current_cost, "Expanded nodes:", expanded_num)

# Save Behavior Tree to file
file_name = "tree"
file_path = f'./{file_name}.btml'
with open(file_path, 'w') as file:
    file.write(ptml_string)

# Load and visualize Behavior Tree
bt = BehaviorTree(file_path, env.behavior_lib)
bt.print()
bt.draw()

# Simulate execution in a simulated scenario
goal = goal_set[0]
print(f"Goal: {goal}")
env.agents[0].bind_bt(bt)
env.reset()

is_finished = False
while not is_finished:
    is_finished = env.step()
    if goal <= env.agents[0].condition_set:
        is_finished = True

env.close()
```

## 🎮 Simulation
The images below illustrate examples from two simulation scenarios, including behavior trees generated by LLM-HOBTEA and their corresponding simulation environments.

### RoboWaiter
The RoboWaiter scenario involves a service robot performing tasks in a café setting.

**Goal:** `On(Coffee,Table3) & Active(AC)`
![](https://raw.githubusercontent.com/DIDS-EI/BTPG/main/images/RW.svg)

### VirtualHome
The VirtualHome scenario simulates a household environment where a robot performs various domestic tasks.

**Goal:** `IsIn(bananas,fridge) & IsClose(fridge)`
![](https://raw.githubusercontent.com/DIDS-EI/BTPG/main/images/VH.svg)

## 📜 License

This project is licensed under the MIT License. See the [LICENSE](LICENSE) file for details.

BTPG

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README