File size: 9,809 Bytes
a71ca3c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 | # :material-silverware-fork-knife: **BEHAVIOR Tasks**
[`BehaviorTask`](../reference/tasks/behavior_task.md) represents a family of **1000 long-horizon household activities** that humans benefit the most from robots' help based on our survey results.
## Getting Started
BDDL (BEHAVIOR Domain Definition Language) is automatically installed when you follow the [standard BEHAVIOR installation guide](../getting_started/installation.md). This section is only needed if you want to customize or edit task definitions.
### Installing BDDL for Customization (Optional)
If you want to modify or create new task definitions, you'll need to install the BDDL repository in development mode:
1. **Clone the repository:**
```bash
git clone https://github.com/StanfordVL/bddl.git
```
2. **Install the package:**
```bash
conda activate behavior
cd bddl
pip install -e .
```
3. **Verify installation:**
```python
>>> import bddl; print(bddl)
<module 'bddl' from '/path/to/BDDL/bddl/__init__.py'>
```
This should now point to your local `bddl` repo, instead of the PyPI version, giving you full editing capabilities.
## Task Library
### Where to Find Tasks
All 1000 activities are defined using **BDDL** (BEHAVIOR Domain Definition Language), specifically designed for household robotics tasks.
- **Local Files:** [`bddl/activity_definitions`](https://github.com/StanfordVL/bddl/tree/master/bddl/activity_definitions) folder
- **Online Browser:** [BEHAVIOR Knowledgebase](https://behavior.stanford.edu/knowledgebase/tasks) for interactive exploration
### Task Structure
Every BEHAVIOR task consists of three main components:
#### 1. Objects
Each line represents a [**WordNet**](https://wordnet.princeton.edu/) synset of required objects.
**Example:**
```yaml
candle.n.01_1 candle.n.01_2 candle.n.01_3 candle.n.01_4 - candle.n.01
```
This means four objects belonging to the `candle.n.01` synset are needed.
#### 2. Initial Conditions
Ground predicates that define the world state when the task begins.
**Example:**
```yaml
(ontop candle.n.01_1 table.n.02_1)
```
The first candle starts on top of the first table.
#### 3. Goal Conditions
Predicates and logical blocks that must be satisfied for task completion.
**Example:**
```yaml
(inside ?candle.n.01 ?wicker_basket.n.01)
```
All candles must end up inside wicker baskets.
### Complete Task Example
Here's a full task definition for assembling gift baskets:
```yaml
(define (problem assembling_gift_baskets-0)
(:domain omnigibson)
(:objects
wicker_basket.n.01_1 wicker_basket.n.01_2 wicker_basket.n.01_3 wicker_basket.n.01_4 - wicker_basket.n.01
floor.n.01_1 - floor.n.01
candle.n.01_1 candle.n.01_2 candle.n.01_3 candle.n.01_4 - candle.n.01
butter_cookie.n.01_1 butter_cookie.n.01_2 butter_cookie.n.01_3 butter_cookie.n.01_4 - butter_cookie.n.01
swiss_cheese.n.01_1 swiss_cheese.n.01_2 swiss_cheese.n.01_3 swiss_cheese.n.01_4 - swiss_cheese.n.01
bow.n.08_1 bow.n.08_2 bow.n.08_3 bow.n.08_4 - bow.n.08
table.n.02_1 table.n.02_2 - table.n.02
agent.n.01_1 - agent.n.01
)
(:init
(ontop wicker_basket.n.01_1 floor.n.01_1)
(ontop wicker_basket.n.01_2 floor.n.01_1)
(ontop wicker_basket.n.01_3 floor.n.01_1)
(ontop wicker_basket.n.01_4 floor.n.01_1)
(ontop candle.n.01_1 table.n.02_1)
(ontop candle.n.01_2 table.n.02_1)
(ontop candle.n.01_3 table.n.02_1)
(ontop candle.n.01_4 table.n.02_1)
(ontop butter_cookie.n.01_1 table.n.02_1)
(ontop butter_cookie.n.01_2 table.n.02_1)
(ontop butter_cookie.n.01_3 table.n.02_1)
(ontop butter_cookie.n.01_4 table.n.02_1)
(ontop swiss_cheese.n.01_1 table.n.02_2)
(ontop swiss_cheese.n.01_2 table.n.02_2)
(ontop swiss_cheese.n.01_3 table.n.02_2)
(ontop swiss_cheese.n.01_4 table.n.02_2)
(ontop bow.n.08_1 table.n.02_2)
(ontop bow.n.08_2 table.n.02_2)
(ontop bow.n.08_3 table.n.02_2)
(ontop bow.n.08_4 table.n.02_2)
(inroom floor.n.01_1 living_room)
(inroom table.n.02_1 living_room)
(inroom table.n.02_2 living_room)
(ontop agent.n.01_1 floor.n.01_1)
)
(:goal
(and
(forpairs
(?wicker_basket.n.01 - wicker_basket.n.01)
(?candle.n.01 - candle.n.01)
(inside ?candle.n.01 ?wicker_basket.n.01)
)
(forpairs
(?wicker_basket.n.01 - wicker_basket.n.01)
(?swiss_cheese.n.01 - swiss_cheese.n.01)
(inside ?swiss_cheese.n.01 ?wicker_basket.n.01)
)
(forpairs
(?wicker_basket.n.01 - wicker_basket.n.01)
(?butter_cookie.n.01 - butter_cookie.n.01)
(inside ?butter_cookie.n.01 ?wicker_basket.n.01)
)
(forpairs
(?wicker_basket.n.01 - wicker_basket.n.01)
(?bow.n.08 - bow.n.08)
(inside ?bow.n.08 ?wicker_basket.n.01)
)
)
)
)
```
## Working with Tasks
### Sampling New Task Instances
Generate fresh instances of existing tasks with randomized elements for variety and robustness testing.
**Example: Sampling a wood floor laying task**
```python
import omnigibson as og
cfg = {
"scene": {
"type": "InteractiveTraversableScene",
"scene_model": "Rs_int",
},
"robots": [
{
"type": "Fetch",
"obs_modalities": ["rgb"],
"default_arm_pose": "diagonal30",
"default_reset_mode": "tuck",
},
],
"task": {
"type": "BehaviorTask",
"activity_name": "laying_wood_floors", # Task name
"activity_definition_id": 0, # Problem variant
"activity_instance_id": 0, # Instance number
"online_object_sampling": True, # Enable sampling
},
}
env = og.Environment(configs=cfg)
```
Each sampling run produces different variations:
- **Object Categories:** High-level synsets sample different specific types (e.g., `fruit.n.01` → apple, banana, orange)
- **Object Models:** Same categories use different 3D models with varying shapes and colors
- **Object Poses:** Spatial arrangements vary while satisfying constraints
### Sampling Challenges
Sampling can fail for various reasons - this is normal behavior:
| Failure Type | Description |
|--------------|-------------|
| **Missing Rooms** | Required room type doesn't exist in the scene |
| **No Valid Objects** | Cannot find appropriate scene objects for the task |
| **Physical Constraints** | Cannot satisfy initial conditions (e.g., objects too large) |
### Saving Task Instances
Once successfully sampled, preserve the configuration for reuse:
```python
# Save the current task instance
env.task.save_task()
```
Default save location:
```
<gm.DATA_PATH>/behavior-1k-assets/scenes/<SCENE_MODEL>/json/<scene_model>_task_{activity_name}_{activity_definition_id}_{activity_instance_id}_template.json
```
## Loading Pre-sampled Tasks
### Using Existing Instances
For consistent, reproducible experiments, load pre-sampled task instances from the dataset. Our publicly available dataset includes **exactly 1 pre-sampled instance** of all 1000 BEHAVIOR tasks.
**Example: Loading a pre-sampled task**
```python
import omnigibson as og
cfg = {
"scene": {
"type": "InteractiveTraversableScene",
"scene_model": "Rs_int",
},
"robots": [
{
"type": "Fetch",
"obs_modalities": ["rgb"],
"default_arm_pose": "diagonal30",
"default_reset_mode": "tuck",
},
],
"task": {
"type": "BehaviorTask",
"activity_name": "laying_wood_floors",
"activity_definition_id": 0,
"activity_instance_id": 0,
"online_object_sampling": False, # Load pre-sampled
},
}
env = og.Environment(configs=cfg)
```
### Finding Pre-sampled Tasks
Discover available pre-sampled task instances in your dataset:
```bash
ls -l <gm.DATA_PATH>/*/scenes/*/json/*task*
```
**Recommendation:** Set `online_object_sampling=False` to load the stable, pre-sampled task instances for consistent evaluation and comparison.
## Creating Custom Tasks
### Step-by-Step Process
1. **Create task directory:**
```bash
mkdir bddl/activity_definitions/my_new_task
```
2. **Create task definition file:**
```bash
touch bddl/activity_definitions/my_new_task/problem0.bddl
```
3. **Define task components:**
- Specify required objects (`:objects`)
- Set initial conditions (`:init`)
- Define goal conditions (`:goal`)
4. **Validate compatibility:**
```bash
cd bddl
python tests/bddl_tests.py batch_verify
python tests/tm_tests.py
```
5. **Test your custom task:**
```python
cfg = {
# ... standard configuration ...
"task": {
"type": "BehaviorTask",
"activity_name": "my_new_task", # Your custom task
"activity_definition_id": 0,
"activity_instance_id": 0,
"online_object_sampling": True,
},
}
env = og.Environment(configs=cfg)
```
### Activity Identification
Tasks are uniquely identified by three parameters:
- **`activity_name`** - The task type (e.g., "laying_wood_floors")
- **`activity_definition_id`** - Problem variant within the task (usually 0)
- **`activity_instance_id`** - Specific instance number for the sampled configuration
You now have the complete toolkit for working with BEHAVIOR tasks! Start by exploring the 1000 pre-defined activities, then create your own custom household robotics challenges. |