| name: ai-executive-assistant |
| version: "1.0" |
| entry_point: env.assistant_env:ExecutiveAssistantEnv |
|
|
| observation_space: |
| type: dict |
| keys: |
| time: |
| type: string |
| description: Current simulation time in HH:MM format |
| tasks: |
| type: list |
| description: List of task objects with id, title, time, duration, priority, type, status |
| inbox: |
| type: list |
| description: List of inbox message objects with id, sender, content, urgency, replied |
| preferences: |
| type: dict |
| description: User preference profile for personalization |
|
|
| action_space: |
| type: discrete |
| actions: |
| - schedule_task |
| - complete_task |
| - defer_task |
| - send_reply |
| - reject_task |
| - ask_clarification |
|
|
| max_steps: 50 |
| reward_range: [-20, 20] |
|
|
| features: |
| temporal_reasoning: true |
| partial_observability: true |
| action_masking: true |
| curriculum_learning: true |
| conflict_graph: true |
| user_preferences: true |
|
|
| description: > |
| RL environment simulating an executive assistant handling scheduling, |
| inbox communication, and task prioritization. Features temporal reasoning |
| with overlap detection, multi-objective reward shaping, partial observability |
| with hidden tasks and delayed inbox, action masking, conflict graph modeling, |
| curriculum learning, and personalization via user preference memory. |
| |
