Breathing Life into the Past: Heterogeneous Crowd Simulation for Digital Heritage

    ## TL;DR
    Researchers have developed a sophisticated methodology to "re-populate" historical sites with dynamic, culturally specific digital crowds. By combining classical flocking algorithms with Hierarchical State Machines and real-time parameter controls, they successfully recreated the bustling 19th-century trading port of George Town, Malaysia, featuring realistic interactions between diverse ethnic groups.

    ## The Missing Link in Digital Heritage
    Most virtual heritage projects excel at building beautiful, but empty, 3D models of ancient cities. To truly understand a site like Weld Quay in George Town, one must see the **interaction**: Indian coolies carrying goods, Malay vendors haggling, and British soldiers patrolling. The "Past Life" is not just about buildings; it is about the **social atmosphere**.

    The challenge is that historical records rarely provide individual-level movement data. This paper addresses this by using **microscopic crowd simulation** to generate emergent behaviors from simple, ethnically-distinct social rules.

    ## Methodology: The Three Pillars of Believable Simulation

    ### 1. Heterogeneous Steering (How they move)
    The system utilizes a particle-based "Boid" algorithm but extends it with **inter-ethnic interaction formalism**. Each ethnic group is assigned specific steering weights for behaviors like alignment, cohesion, and separation. 
    *   **Malay Sellers**: Exhibit "Seeking" and "Wandering" behaviors to find customers.
    *   **Indian Coolies**: Use "Queuing" and "Path-following" to load/unload ships.
    *   **British Soldiers**: Utilize "Leader-following" for disciplined patrols.

    ### 2. Hierarchical State Machine (What they do)
    To handle complex logic (e.g., a thief deciding to rob a vendor), the authors implemented a **Hierarchical State Machine (HSM)**. This allows for individual "storylines." An agent might be in a "Wandering" state, shift to "Stealing" if the "isPoliceNearby" condition is false, and then transition to "Being Caught" and "Following Police."

    ![Model Architecture of State Transitions](https://cdn.atominnolab.com/wisdoc/images/20260526-ad5d63f4-31c8-43c0-9565-eac6b0bf60ad/page_013_block_006.png)
    *Fig 1. Inter-ethnic interaction models showing global and local event triggers.*

    ### 3. High-Level Controller (Real-time Adaptation)
    A custom "Flock Manager" allows designers to tweak parameters—like speed, mass, and vision angle—in real-time during the simulation. This is vital for "look-and-feel" validation where mathematical precision is secondary to visual believability.

    ## Strategic Experimental Results
    The methodology was tested in a reconstructed 19th-century George Town environment. By introducing **Mass as a dynamic variable** (e.g., Indian coolies becoming 200% heavier when carrying goods), the simulation achieved realistic acceleration and momentum changes.

    ![Experimental Result: Interaction Scenarios](https://cdn.atominnolab.com/wisdoc/images/20260526-ad5d63f4-31c8-43c0-9565-eac6b0bf60ad/page_022_block_005.png)
    *Fig 2. Validation of specific social interactions: A British captain greeting a client while his soldiers follow in formation.*

    Key Achievements:
    - **Scale**: Real-time simulation of heterogeneous groups ranging from 200 to 20,000 agents.
    - **Complex Formations**: Successful "V-like" and "Line-Abreast" formations for military parades and escorting missions.
    - **Navigation Utility**: A new "Smooth Turning" algorithm based on angle prediction prevented the common "deadlock" problem where agents get stuck at narrow pier corners.

    ## Critical Insight: Why This Matters
    The paper highlights a shift from **Tangible Heritage** (buildings) to **Intangible Heritage** (behaviors). By formalizing social interactions as computable states, the authors provide a template for "Virtual Museums" where visitors don't just see the past—they participate in its rhythm.

    ### Limitations & Future Work
    While the "look-and-feel" is strong, the model lacks a **Chaotic System** or **Economic Model** to drive emergent behaviors based on resource scarcity or fatigue. Future iterations plan to integrate "Crowd Patches" for distant, more deterministic crowds to save computational resources while focusing real-time microscopic simulation on agents near the user.

    ## Conclusion
    This methodology proves that digital heritage is most effective when it is **alive**. Through the lens of 19th-century Malaysia, we see how ethnic diversity informs the global motion of a city, providing a more profound historical narrative than a static model ever could.