Continuous Self-Calibration for Lifelong Autonomy

任何多传感器机器人系统的基本要求是精确校准。为了确保最佳性能，传感器需要在内部和相互之间进行适当校准。不正确的校准会导致错误的数据融合，在最坏的情况下，可能会导致潜在的危险操作或导航故障。如今，机器人操作和导航技术通常依赖于离线校准程序，通常由人类操作员执行。校准过程通常繁琐且耗时，每当传感器重新定位或施加大量机械应力时，必须重复该过程。这种对校准的需求严重限制了我们建造能够长时间独立工作的“通电即走”机器人的能力。拟议的研究将开发能够在正常操作期间更新和维护自身校准的机器人系统，而无需使用外部参考、设备或协助。这将涉及（i）建立统一的数学模型用于多种异质传感器的联合校准的（贝叶斯）框架（ii）设计方法以在该框架内确定在给定运动和其他约束的情况下可以精确地校准哪些系统参数，（iii）设计平台轨迹，以准确和快速地校准传感器，以及（iv）演示在几天到几周的时间内成功自动校准一系列自主平台上的多个传感器。这项工作的一个重要方面是严格评估和验证在非结构化和动态环境（例如，穿越崎岖地形的行星漫游车）。该研究将支持培训熟悉最先进传感器系统及其操作和校准的高素质人员。反过来，这项工作将大大有助于发展加拿大在不断增长的机器人领域的专业知识。

A fundamental requirement in any multisensor robotic system is precision calibration. To ensure optimal performance, the sensors need to be properly calibrated, both internally and relative to one another. Incorrect calibration results in erroneous data fusion and, in the worst case, may lead to potentially dangerous manipulation or navigation failures. Today, robotic manipulation and navigation techniques usually depend on offline calibration procedures, typically carried out by a human operator. The calibration process, which is often tedious and time consuming, must be repeated whenever sensors are repositioned or substantial mechanical stresses are applied. This need for calibration severely limits our ability to build 'power-on-and-go' robots that are able to function independently for long durations. The proposed research will develop robotic systems that are able to update and maintain their own calibration during normal operation, without the use of external references, equipment, or assistance. This will involve (i) establishing a unified mathematical (Bayesian) framework for joint calibration of a variety of heterogenous sensors (including inertial, odometric, encoder, visual, LIDAR, and RGB-Depth modalities), (ii) devising methods to determine, within this framework, exactly which system parameters can be calibrated, given motion and other constraints, (iii) designing platform trajectories to calibrate the sensors accurately and rapidly, and (iv) demonstrating successful automatic calibration of multiple sensors on-board a range of autonomous platforms, over periods of days to weeks. An important aspect of the work is the rigorous assessment and verification of the performance of the calibration algorithms on deployed field robots operating in unstructured and dynamic environments (e.g., planetary rovers traversing rugged terrain). The research will support the training of highly qualified personnel who are familiar with state-of-the-art sensor systems, their operation and their calibration. In turn, the work will contribute substantially to developing Canadian expertise in the growing robotics field.