# prob-ff

**Repository Path**: xiaozq22/prob-ff

## Basic Information

- **Project Name**: prob-ff
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2025-05-15
- **Last Updated**: 2025-05-17

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# 基于概率规划的行为树生成算法设计及实现

## 项目下载
```
git clone https://gitee.com/xiaozq22/prob-ff.git
cd prob-ff
```

**项目克隆下来后首先要将Cachet-1.21-wmc项目中的cachet文件拷贝到`~/CACHET/`目录下，并为其添加可执行权限**
```
mkdir -p ~/CACHET
cp ./Cachet-1.21-wmc/cachet ~/CACHET/
chmod +x ~/CACHET/cachet
```

## 编译及运行

1. 编译
```
cd <当前目录>
make
```
如果不报错的话，会在当前目录下生成名为`ff`的可执行文件，接着可以运行如下示例进行检验：

2. 运行
以`tests`目录中的block任务为例，domain文件如下：
```
(define (domain blocksworld)
    (:predicates
        (clear ?x)
        (on-table ?x)
        (on ?x ?y)
    )

    (:action move-b-to-b
        :parameters (?bm ?bf ?bt)
        :precondition (and (clear ?bm) (clear ?bt) (on ?bm ?bf))
        :effect (and (not (clear ?bt)) (not (on ?bm ?bf))
            (on ?bm ?bt) (clear ?bf))
    )

    (:action move-bstack-to-t
        :parameters (?b ?bl)
        :effect (and (when
                (on ?b ?bl)
                (and (not (on ?b ?bl)) (on-table ?b) (clear ?bl))))
    )

    (:action move-t-to-b
        :parameters (?bm ?bt)
        :effect (and (when
                (and (clear ?bt) (on-table ?bm))
                (and (not (clear ?bt)) (not (on-table ?bm))
                    (on ?bm ?bt))))
    )
)
```

表示可执行的动作有`move-b-to-b`（将物块从一个物块移动到另一个物块上）, `move-bstack-to-t`（将物块从一个物块移动到桌子上）, `move-t-to-b`（将物块从桌子移动到另一个五块上）三个。

problem文件定义如下：
```
(define (problem BW-simple-3)
    (:domain blocksworld)
    (:objects
        b1 b2
    )

    (:init
        ;; 核心确定性事实
        (on-table b1)
        (clear b1)

        ;; 概率化初始状态配置
        (cpt
            (on-table b2) 0.6) ; b2有60%概率在table上
        (cpt
            (clear b2) 0.6) ; b2有60%概率clear
    )

    (:goal 0.6 ; 目标状态概率
        (and
            (on b1 b2) ; 目标堆叠结构，b1位于b2上
            (clear b1) ; b1顶部保持clear
        )
    )
)
```

初始状态下，b1位于桌面上，b2有60%的可能性位于桌面上；目标状态下，需要有至少60%的概率实现：b1位于b2上。

使用如下命令运行：
```
./ff -o tests/block_domain.pddl -f tests/block_problem_2.pddl
```

可以得到如下输出结果：
```
ff: parsing domain file
domain 'BLOCKSWORLD' defined
 ... done.
ff: parsing problem file
problem 'BW-SIMPLE-2' defined
 ... done.


initial BN is cross-product of independent multi-state vars! enabling special case treatment!

Cueing down from goal distance:    2 into depth [1]
                                   0            

ff: found legal plan as follows

step    0: MOVE-T-TO-B B1 B2
     

statistics:    0.00 seconds instantiating 16 easy, 0 hard action templates
               0.00 seconds reachability analysis, yielding 16 facts and 16 actions
               0.00 seconds creating final representation with 16 relevant 16 final facts (16 max U, 8 CNF max U)
               0.00 seconds building connectivity graph
               0.00 pure seconds (   0.00 combined) evaluating 2 states, to a max depth of 1
               0.00 sec (   0.01 setup), 1 calls to external Cachet, 0 lits removed in 0 WMC calls, 1 ``fast-simulated'' calls
               0.00 seconds in DP for 0 RPG ini state implication checks
               0.00 seconds in DP for 0 RPlan extract ini state implication checks (0 lits removed)
               0.00 seconds generating,    0.00 seconds encoding 1 state transition base CNFs
               0.00 seconds in DP solving 3 state transition CNFs
               0.00 seconds checking for stagnating states (0 hits), including 0 DP calls
                  3 total DP calls, 0 total UP calls,    0.00 sec membership
               0.00 seconds for remaining searching duties
               0.01 seconds total time (+ 0.00 secs for CNF-SAT communication; + 0.00 secs for CNF-WMC communication; + 0.00 secs for CNF memory allocation; there may be further overhead for Cachet WMC-CNF parsing, which is not included, and/or for file creation times)
```

观察上述结果可知：为了实现60%的成功概率，只需执行一步（ff: found legal plan as follows后面内容）：`MOVE-T-TO-B B1 B2`即可。

3. 将结果转化为行为树（以[BehaviorTree.CPP](https://www.behaviortree.dev/docs/learn-the-basics/xml_format)给出的xml格式为例）：
```
 <root BTCPP_format="4">
     <BehaviorTree ID="MainTree">
        <Sequence name="root_sequence">
            <MOVE-T-TO-B    name="MOVE-T-TO-B B1 B2"/>
        </Sequence>
     </BehaviorTree>
 </root>
 ```

画成图，即为一个`Sequence`节点下面接一个名为`MOVE-T-TO-B B1 B2`的动作节点。