本文为美国卡内基梅隆大学（作者：Anirudh Vemula）的硕士论文，共66页。

为了让移动机器人变得无处不在，它们需要能够在动态环境中安全高效地导航。这是一个具有挑战性的问题，因为搜索空间中增加了时间维度，而且动态智能体之间的微妙交互非常难以建模。

在本文中，我们将解决这两个挑战。提出了一种新的路径规划算法，解决了动态智能体环境下的维数问题。在不考虑机器人动力学模型的前提下，应用自适应维数的思想，加快了机器人在动态环境中的路径规划速度。具体来说，我们的方法只在可能发生碰撞的环境区域中考虑时间维度，而在其他区域的低维状态空间中进行规划。我们证明了所用方法是完备的，并且保证在一个成本次优界内找到一个解。通过对动态环境下三维非完整车辆导航问题的实验验证了该方法的有效性。结果表明，该方法在基于4D启发式算法A*的规划时间上取得了显著提高，特别是当所得到的规划与启发式算法所建议的规划有较大的偏差时。

我们提出了一种新的统计模型来捕捉群体中的合作行为，从而解决了建模交互的挑战。以前的方法都是使用基于接近度的手工函数来模拟人-人和人-机器人的交互。然而，这些方法只能对简单的交互进行建模，不能推广到复杂的拥挤环境中。我们开发了一种方法，通过使用真实人体轨迹数据训练的局部交互模型，对群体中所有交互主体未来轨迹的联合分布进行建模。交互模型根据附近其他智能体的空间方向推断每个智能体的运动速度。在预测过程中，我们的方法从智能体过去的轨迹中推断出运动的目标，并利用学习到的模型来预测智能体未来的轨迹。我们在一个公共数据集上展示了我们的方法相对于最先进方法的性能，并且表明我们的模型在预测未来较长时间内的轨迹时具有更好的性能。最后，提出了动态环境下机器人导航领域未来的研究方向和面临的挑战。我们计划在一个放置在真实群体中的机器人身上验证本文所提出的工作。其他挑战包括更精确的长期预测、与预测相关的不确定性和实时增量规划算法。

For mobile robots to become ubiquitous,they need to be able to navigate in dynamic environments in a safe andefficient way. This is a challenging problem due to the added time dimension inthe search space and the subtle interactions between dynamic agents that areextremely difficult to model. In this thesis, we will address both of thesechallenges. The challenge of dimensionality is addressed by proposing a novelpath planning algorithm in environments with dynamic agents with quick planningtimes. We apply the idea of adaptive dimensionality to speed up path planningin dynamic environments for a robot with no assumptions on its dynamic model.Specifically, our approach considers the time dimension only in those regionsof the environment where a potential collision may occur, and plans in alow-dimensional state-space elsewhere. We show that our approach is completeand is guaranteed to find a solution, if one exists, within a costsub-optimality bound. We experimentally validate our method on the problem of3D nonholonomic vehicle navigation in dynamic environments. Our results showthat the presented approach achieves substantial speedups in planning time over4D heuristic-based A*, especially when the resulting plan deviatessignificantly from the one suggested by the heuristic. We tackle the challengeof modeling interactions by presenting a novel statistical model to capturecooperative behavior in human crowds. Previous approaches have used handcraftedfunctions based on proximity to model human-human and human-robot interactions.However, these approaches can only model simple interactions and fail togeneralize for complex crowded settings. We develop an approach that models thejoint distribution over future trajectories of all interacting agents in thecrowd through a local interaction model that we train using real humantrajectory data. The interaction model infers the velocity of each agent basedon the spatial orientation of other agents in his vicinity. During prediction,our approach infers the goal of the agent from its past trajectory and uses thelearned model to predict its future trajectory. We demonstrate the performanceof our method against a state-of-the-art approach on a public dataset and showthat our model outperforms when predicting future trajectories for longerhorizons. Finally, we lay out future directions of research in the domain ofrobot navigation in dynamic environments, and the challenges remaining. We planto verify and validate the proposed work in this thesis on a robot placed in areal human crowd. Other challenges include more accurate long-term prediction,uncertainty associated with predictions and real-time incremental planningalgorithms.

1.  引言

2. 拥挤人群中的导航：项目调查

3. 基于自适应维数的动态环境路径规划

4. 拥挤人群中协作导航的建模

5. 结论

附录A 代码与发表的论文

附录B 公开数据集

更多精彩文章请关注公众号：