Back chaining is a goal-directed way to construct an autonomous behavior tree (ABT): begin with the desired behavior, then work backward to identify actions, conditions, communication patterns, and the required tree structure. section 6.5 section 6.7
A practical back-chaining workflow is:
- Identify the goal behavior the agent must perform. section 6.5
- Work backward to determine the actions that lead to the goal. section 6.5
- Specify conditions required for each action to succeed. section 6.5
- Choose how assistants communicate, then add action and condition nodes recursively until they connect to known states or facts. section 6.5
Building an assistant
For the goal “Create an assistant that can help me do {task},” the required actions are to create, verify, test, name, and instruct the assistant. These become nodes in an ABT, with verification represented as a condition. Assistants may share a message thread, while blackboards or files can communicate information between nodes. section 6.5
Listing 6.11 ABT for building an assistant
Root │ ├── Sequence │ ├── Action: Give the assistant relevant instructions to help a user with a given task │ ├── Action: Name the assistant │ ├── Action: Test the assistant │ ├── Condition: Verify the assistant │ └── Action: Create the assistant
Inspect the assistant-building ABT as a compact tree showing its root, sequence, actions, and verification condition. section 6.5
