Planning Beyond The Sensing Horizon Using a Learned Context

Abstract

Last-mile delivery systems commonly propose the use of autonomous robotic vehicles to increase scalability and efficiency. The economic inefficiency of collecting accurate prior maps for navigation motivates the use of planning algorithms that operate in unmapped environments. However, these algorithms typically waste time exploring regions that are unlikely to contain the delivery destination. Context is key information about structured environments that could guide exploration toward the unknown goal location, but the abstract idea is difficult to quantify for use in a planning algorithm. Some approaches specifically consider contextual relationships between objects, but would perform poorly in object-sparse environments like outdoors. Recent deep learningbased approaches consider context too generally, making training/transferability difficult. Therefore, this work proposes a novel formulation of utilizing context for planning as an image-to-image translation problem, which is shown to extract terrain context from semantic gridmaps, into a metric that an exploration-based planner can use. The proposed framework has the benefit of training on a static dataset instead of requiring a time-consuming simulator. Across 42 test houses with layouts from satellite images, the trained algorithm enables a robot to reach its goal 189% faster than with a context-unaware planner, and within 63% of the optimal path computed with a prior map. The proposed algorithm is also implemented on a vehicle with a forward-facing camera in a high-fidelity, Unreal simulation of neighborhood houses.

Waypoint Planning Networks

“…The map module has both Convolutional Auto-encoder (CAE) [25], similar to [45], and LSTM components and attempts to fix some issues, e.g. greediness, that are present in the view module (i.e.…”

Online High-Level Model Estimation for Efficient Hierarchical Robot Navigation

“…While hierarchical planning can be efficient, it is obvious that the quality of the high-level planning problem, or the combination of A, c(a; Γ), and f(a; Γ), dictates the ability of the high-level search process to effectively guide primitive planning. Intuitively, a good choice of representation is one that ensures plan similarity 2 , or that low-cost, feasible plans in P have low-cost, feasible primitive refinements in P. Unfortunately, generating planning representations which exhibit plan similarity can be challenging, and in practice such representations are often hand-designed for specific problems [24], or require significant prior domain knowledge to be generated [12], [15], [16]. Instead, we define a simple A and use online planning results to infer estimates of c(a; Γ) and f(a; Γ) over the course of a multi-query planning trial.…”

“…Other approaches use techniques such as wavelet decompositions [13] or the information bottleneck [14] to compress known cost maps for multiresolution planning, but do not explicitly capture the structure of the paths agents use when traversing in the environment. Recently, some approaches have turned to learning to generate coarse representations of planning problems based on offline datasets of primitive plans and additional environmental cues, such as semantically segmented overhead images [15] or height maps [16]; however, such additional environmental cues can be challenging to acquire in real-world environments.…”