Project Overview
GT RoboRambler is Georgia Tech’s RoboMaster competition team, building autonomous combat robots for the annual RoboMaster University Championship. As Software Team Lead, I’m responsible for the autonomy stack of the team’s Sentry Robot.
Key Contributions
- Autonomous decision system: Architected a BehaviorTree-based autonomous decision system for the Sentry Robot, structured around five plugin nodes and a global Blackboard for shared state across behaviors.
- Chassis power regression: Modeled a two-stage, physics-informed chassis power regression from system identification data, enabling real-time power limiting under competition constraints.
- Localization pipeline: Built a localization pipeline combining AprilTag-based initialization with LiDAR point cloud relocalization on a direct time-of-flight (dToF) sensor.
Why BehaviorTree.CPP
Behavior trees offer a modular, debuggable alternative to monolithic state machines for robot decision-making. Structuring the Sentry Robot’s autonomy around plugin nodes and a shared Blackboard makes it straightforward to add new behaviors — such as target prioritization or retreat conditions — without rewriting the existing control logic, which matters during the rapid iteration cycles of competition prep.
Power-Aware Chassis Control
RoboMaster competition rules cap total chassis power draw, so accurately predicting power consumption from motor commands is critical to staying within limits while maximizing performance. The two-stage regression is fit directly on system identification data collected from the robot, allowing the power limiter to react in real time as conditions change during a match.
Localization
Reliable localization is a prerequisite for any higher-level autonomy. The pipeline combines a global AprilTag-based initialization step with continuous LiDAR point cloud relocalization, giving the Sentry Robot a robust position estimate even as it moves around the field.