Introduction
Behavior Trees (BTs) have become a standard for modular task switching in robotics due to their composability and readability. However, the standard "Parallel" node often fails in real-world scenarios where sub-behaviors must progress at coordinated rates. Without synchronization, a robot's head might finish a scanning motion before its base reaches the target, leading to failed task execution or unnatural movements. This paper introduces two synchronization mechanisms—Absolute and Relative—to enforce progress constraints in parallel BT executions.
Proposed Synchronization Nodes
The authors propose extending the classical BT framework with two novel control flow nodes designed to handle concurrency issues:
Absolute Synchronized Parallel Node ($\Rightarrow^A$)
This node coordinates children using a set of discrete "barriers" representing progress percentages (e.g., 25%, 50%, 75%). A child node that reaches a barrier is paused—it stops receiving activation signals or "ticks"—until all other children have reached the same barrier. This ensures that no task completes a major phase significantly ahead of others.
Relative Synchronized Parallel Node ($\Rightarrow^R$)
Instead of fixed points, this node utilizes a threshold $\Delta$. It ensures that no child node's progress exceeds the slowest child's progress by more than the amount $\Delta$. If the gap exceeds this threshold, the faster node is throttled until the others catch up.
Experimental Evaluation
The proposed nodes were evaluated through 1,000-run Monte Carlo simulations and practical robot experiments using a simulated R1 humanoid. The primary metric used was the Synchronization Distance ($\pi$), which quantifies the progress disparity between parallel children. Simulations analyzed how parameters like the number of barriers, threshold values, and execution noise ($\omega$) affected performance.
Key Results
Simulations demonstrated that both mechanisms effectively reduce the synchronization distance. For the Absolute node, increasing the number of barriers ($|B|$) leads to significantly tighter coordination, even as execution noise increases.
In robot experiments, these nodes allowed an R1 robot to successfully perform "Object Seeking" by throttling head movement to match base movement. In the unsynchronized version, the head moved too quickly, causing the robot to miss objects entirely. Furthermore, by including an "artificial" child node representing an ideal timeline, the system could force the robot to follow specific velocity profiles (e.g., sigmoid or linear) with high predictability.
Discussion and Limitations
While synchronization improves coordination, it introduces a trade-off: intermittency. Tight constraints (many barriers or small $\Delta$) can cause "stuttering" or jerky movements as nodes frequently pause to wait for laggards. Additionally, the overall execution speed is always limited by the slowest sub-task. The paper suggests that future work should focus on automating the placement of barriers based on robot dynamics to minimize these effects.
Conclusion
By introducing Absolute and Relative synchronization, this work makes parallel execution in BTs safe and predictable. These nodes bridge the gap between high-level task composition and low-level execution constraints, enabling more complex and reliable robotic behaviors.
Summary updated