SmartCam 多传感器机器人SmartCam · RGB-D Robot Perception
我在 Rice ELEC 594 团队中负责 RGB-D 感知，以及 Gazebo 规则示教、数据转换与 SmolVLA 离线实验，连接传感输入、机器人数据和动作预测。In the Rice ELEC 594 team, my work spans RGB-D perception, Gazebo rule-based demonstrations, dataset conversion and SmolVLA offline experiments, connecting sensing, robot data and action prediction.
2025.08—至今 · 个人感知与仿真工作 / 团队系统集成Aug 2025–present · Personal perception and simulation / team system integration
我完成的四项核心工作Four core contributions
RGB-D 目标感知RGB-D perception
RealSense 对齐、YOLO 检测、有效深度中值与相机 XYZ 输出。RealSense alignment, YOLO detection, valid-depth medians and camera XYZ output.
AutoTeacher 规则示教AutoTeacher demonstrations
Gazebo 视觉跟踪、限幅速度控制与带任务标签的轨迹记录。Gazebo visual tracking, bounded velocity control and task-labeled trajectories.
LeRobot 数据转换LeRobot conversion
审核 ROS bag，按时间戳对齐图像、里程计和二维速度动作。Audit ROS bags and align images, odometry and planar velocity actions by timestamp.
SmolVLA 离线实验SmolVLA offline experiments
100 步 BF16 流程测试、Checkpoint 重载与离线动作误差分析。100-step BF16 pipeline test, checkpoint reload and offline action-error analysis.
成果与完成阶段Results and current stage
已实现目标类别、距离与相机 XYZ 输出，打通规则示教到 LeRobot 数据转换流程，并完成 100 步 BF16 流程测试、Checkpoint 重载及离线动作预测。Implemented object-class, distance and camera XYZ output; connected rule-based demonstrations to LeRobot conversion; completed a 100-step BF16 pipeline test, checkpoint reload and offline action prediction.
VLA 当前处于有限仿真数据和离线验证阶段；机器人平台、SLAM、AEB 与 RL 实验归属团队。详情页提供实现、图像、录像、代码解析和完整验证范围。VLA is at the limited-simulation-data and offline-validation stage. The robot platform, SLAM, AEB and RL experiments are team work. The detail page includes implementation, images, videos, code analysis and evaluation scope.
查看完整工程详情 →Explore the full engineering case study →