Data-driven planning via imitation learning

Abstract

Robot planning is the process of selecting a sequence of actions that optimize for a task specific objective. For instance, the objective for a navigation task would be to find collision free paths, while the objective for an exploration task would be to map unknown areas. The optimal solutions to such tasks are heavily influenced by the implicit structure in the environment, i.e. the configuration of objects in the world. State-of-the-art planning approaches, however, do not exploit this structure, thereby expending valuable effort searching the action space instead of focusing on potentially good actions. In this paper, we address the problem of enabling planners to adapt their search strategies by inferring such good actions in an efficient manner using only the information uncovered by the search up until that time.We formulate this as a problem of sequential decision making under uncertainty where at a given iteration a planning policy must map the state of the search to a planning action. Unfortunately, the training process for such partial information based policies is slow to converge and susceptible to poor local minima. Our key insight is that if we could fully observe the underlying world map, we would easily be able to disambiguate between good and bad actions. We hence present a novel datadriven imitation learning framework to efficiently train planning policies by imitating a clairvoyant oracle -an oracle that at train time has full knowledge about the world map and can compute optimal decisions. We leverage the fact that for planning problems, such oracles can be efficiently computed and derive performance guarantees for the learnt policy. We examine two important domains that rely on partial information based policies -informative path planning and search based motion planning. We validate the approach on a spectrum of environments for both problem domains, including experiments on a real UAV, and show that the learnt policy consistently outperforms stateof-the-art algorithms. Our framework is able to train policies that achieve upto 39% more reward than state-of-the art information gathering heuristics and a 70x speedup as compared to A* on search based planning problems. Our approach paves the way forward for applying data-driven techniques to other such problem domains under the umbrella of robot planning.

FLYBO: A Unified Benchmark Environment for Autonomous Flying Robots

“…Given the structured, unified nature of FLYBO, the proposed data and framework can be used to support the development of Reinforcement-Learning (RL) based approaches to autonomous exploration. Methods that build on RL typically require a moderate amount of training data [13] and an efficient means of scaling the number of experiments which could naturally benefit from the flexibility of our system.…”

Leveraging Experience in Lazy Search

“…The realizability gap between the two is vast, resulting in a trivially large regret bound [14]. Instead, Choudhury et al [7] show that imitating the clairvoyant oracle is in fact equivalent to imitating a corresponding hallucinating oracle, that computes an instantaneous posterior over worlds given the edge evaluations so far and computes the expected clairvoyant oracle action value over this posterior i.e,…”

“…Interestingly, if we were to reveal the status of all the edges during training, we can conceive of a clairvoyant oracle [7] that can select the optimal sequence of edges to invalidate. In fact, we show that the oracular selector is equivalent to set cover, for which greedy approximations exist.…”

Learning to Predict Ego-Vehicle Poses for Sampling-Based Nonholonomic Motion Planning

“…It has to be noted that, similar to [12], motion planning for data generation is conducted with full environmental information in order to guarantee fast convergence to a cost-minimizing solution. In contrast to that, the recorded occupancy grid only fuses the current history of sensor observations resulting in unobserved areas due to occlusions.…”