Perceptive whole-body planning of highly maneuverable robots in confined spaces

生物系统(如鸟类、狗)可以自适应地利用肢体间的协调进行运动，以处理不同的情况，如飞行通过不同的几何形状(如小开口)或在受限空间内爬行。神经生理学研究表明，适应性协调来自神经活动、可塑性、肌肉骨骼系统和环境的动态相互作用。最近开发的腿式机器人和高度机动的飞机非常多才多艺，具有复杂的控制机制，使它们能够移动和调整自己的运动。然而，对于自主机器人系统来说，在3D(三维)复杂的受限空间内实现有效和灵活的机动仍然是一个挑战。由于与这种机器人相关的计算复杂性，没有在线计划者在狭小的空间中进行感知的全身运动。在这个项目中，将为具有多个自由度的机器人(即多腿机器人和高度机动的无人机)开发新的感知规划方法，这些方法将生成3D身体姿势和相关的立足点或飞行轨迹，以避免碰撞。来自机载传感器的测量将创建机器人周围环境的3D地图。将开发一种统一的机动机制，将机器人的运动学和动态运动能力与对环境和机器人可能遇到的干扰的知识相结合。该方法将针对革命性的灵活多腿机器人和高度机动的自主飞机，由PI开发的这一拨款申请。该方法将随机采样身体姿势，然后平滑得到的轨迹，同时满足几个约束，如机器人运动和任务最大时间。腿部机器人的立足点和腿部摆动轨迹，以及空中机器人的推力矢量运动，将基于3D环境和机器人与其交互来计算。机器人的身体姿势将得到优化，以确保在有限空间内的稳定机动行为(例如，在倒塌的结构中向下移动)，并在应对任何干扰时避免碰撞。设想的全身规划将被开发成在真实的机器人平台上在线运行，并生成几米长的运动轨迹-足以在有限的空间内逐步移动，在那里只可能感知到部分环境。这些发展将在不同的模拟中进行分析，并通过两足、四足和倾斜旋翼机器人进行实验测试，以演示对不同机器人配置的适用性。测试将在加拿大紧急救援培训设施提供的现实受限空间场景中进行。对科学/工业/加拿大的重要性是增加机器人部署的新范例，这将对使用自主系统的加拿大部门产生重大经济影响。

Biological systems (e.g. birds, dogs) can adaptively use their interlimb coordination for locomotion to deal with different situations such as flying through varying geometries (e.g. small openings) or crawling inside confined spaces. Neurophysiological studies have revealed that the adaptive coordination emerges from dynamic interactions of neural activities, plasticity, musculoskeletal systems, and the environment. Recently-developed legged robots and highly maneuverable aircraft are exceedingly versatile, having sophisticated control mechanisms allowing them to move and adapt their locomotion. Nevertheless, for autonomous robotic systems, achieving effective and agile maneuvering inside 3D (three-dimensional) complex confined spaces remains a challenge. Because of the computational complexity associated with such robots, no online planners exist for perceptive whole-body locomotion in tight spaces. In this project, new methods for perceptive planning will be developed for robots having multi-degrees of freedom (i.e. multi-legged robots and highly maneuverable drones), which will generate 3D body poses and the associated footholds or flight trajectories for collision avoidance. Measurements from onboard sensors will create a 3D map of the environment around the robot. A unified maneuvering mechanism combining the robot's kinematic and dynamic motion capabilities with knowledge about the environment and disturbances that the robot might encounter will be developed. The approach will target revolutionary agile multi-legged robots and highly maneuverable autonomous aircraft developed by the PI of this grant application. The approach will randomly sample body poses, then smooth the resulting trajectory while satisfying several constraints, such as robot motion and mission maximum time. Footholds and legged swing trajectories for legged robots, and thrust vectoring motions for aerial robots, will be computed based on the 3D environment and the robot's interaction with it. The robot's body pose will be optimized to ensure stable maneuvering behaviours inside confined spaces (e.g., moving down inside a collapsed structure), and collision avoidance while coping with any disturbances encountered. The envisioned whole-body planning will be developed to run online on real robot platforms and generate motion trajectories several meters long - sufficient to progressively move inside confined spaces where perception of only part of the environment is possible. The developments will be analyzed in diverse simulations and experimentally tested via biped, quadruped, and tilt-rotorcraft robots to demonstrate applicability to different robot configurations. Tests will be performed in realistic confined-space scenarios available at emergency rescue training facilities in Canada. The importance for science / industry / Canada are new paradigms for increase robot deployment which will have significant economic impact on Canada's sectors using autonomous systems.