Bayesian Learning for Safe High-Speed Navigation in Unknown Environments

Richter, Charles; Vega-Brown, William; Roy, Nicholas

doi:10.1007/978-3-319-60916-4_19

Cited by 55 publications

(52 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…And second, a prediction based on training data can account for characteristics of the environment that are not explicitly visible in the image but are implied by the visual appearance, such as free space around a blind corner. Similar to our previous work in Richter et al [29], our objective in training a model to predict collision probabilities is to capture these advantages implicitly through training examples. We approach this problem as a probabilistic binary classification problem: Given a camera image, and some choice of action, what is the probability that a future collision is inevitable beyond the end of the chosen action?…”

Section: Learning To Predict Collisionsmentioning

confidence: 98%

“…In our prior work, we considered model uncertainty in collision prediction, but focused on features of geometric maps, rather than images and neural networks [29]. Grimmett et al [12] suggest that if a classification error is to be made in robotic decision making, it should be made with high reported uncertainty so that the robot can avoid consequences of a wrong decision.…”

Section: Related Workmentioning

confidence: 99%

“…In the next section, we will motivate and present the problem formulation of our robot navigation scenario, which builds upon our prior work in Richter et al [29]. Then, we will describe our method of learning to predict collisions from camera images, and our approach to novelty detection.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Safe Visual Navigation via Deep Learning and Novelty Detection

Richter

Roy

2017

Robotics: Science and Systems XIII

Self Cite

173

174

View full text Add to dashboard Cite

Abstract-Robots that use learned perceptual models in the real world must be able to safely handle cases where they are forced to make decisions in scenarios that are unlike any of their training examples. However, state-of-the-art deep learning methods are known to produce erratic or unsafe predictions when faced with novel inputs. Furthermore, recent ensemble, bootstrap and dropout methods for quantifying neural network uncertainty may not efficiently provide accurate uncertainty estimates when queried with inputs that are very different from their training data. Rather than unconditionally trusting the predictions of a neural network for unpredictable real-world data, we use an autoencoder to recognize when a query is novel, and revert to a safe prior behavior. With this capability, we can deploy an autonomous deep learning system in arbitrary environments, without concern for whether it has received the appropriate training. We demonstrate our method with a vision-guided robot that can leverage its deep neural network to navigate 50% faster than a safe baseline policy in familiar types of environments, while reverting to the prior behavior in novel environments so that it can safely collect additional training data and continually improve. A video illustrating our approach is available at: http://groups.csail.mit.edu/rrg/videos/safe visual navigation.

show abstract

Section: Learning To Predict Collisionsmentioning

confidence: 98%

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Safe Visual Navigation via Deep Learning and Novelty Detection

Richter

Roy

2017

Robotics: Science and Systems XIII

Self Cite

173

174

View full text Add to dashboard Cite

show abstract

“…Second, by leveraging the theoretical benchmark and combining the aforementioned notions of re-planning, ICS, and forward-looking biasing in a self-contained framework, we derive a pseudo-optimal class of policies that can seamlessly incorporate any amount of prior or learned information while still guaranteeing the robot never collides. Finally, we demonstrate the practicality of our algorithmic approach in numerical experiments using a range of environment types and robot dynamics, including a comparison with a state of the art method for planning in unknown environments [22]. Importantly, computation times are on the order of 0.5-1s of (serial) computation time per action, making it an algorithm amenable to real-time implementation.…”

Section: Introductionmentioning

confidence: 99%

“…More in general, one can frame the problem of planning in unknown environments as a partially observable Markov decision process, where the partial observability is with respect to the environment [22]. To make the problem tractable, one needs to consider a number of approximations, for example, replacing collision avoidance constraints with penalties on collision probabilities (possibly learned [22]). …”

Section: Introductionmentioning

confidence: 99%

Safe Motion Planning in Unknown Environments: Optimality Benchmarks and Tractable Policies

Janson¹,

Hu²,

Pavone³

2018

Robotics: Science and Systems XIV

View full text Add to dashboard Cite

Abstract-This paper addresses the problem of planning a safe (i.e., collision-free) trajectory from an initial state to a goal region when the obstacle space is a-priori unknown and is incrementally revealed online, e.g., through line-of-sight perception. Despite its ubiquitous nature, this formulation of motion planning has received relatively little theoretical investigation, as opposed to the setup where the environment is assumed known. A fundamental challenge is that, unlike motion planning with known obstacles, it is not even clear what an optimal policy to strive for is. Our contribution is threefold. First, we present a notion of optimality for safe planning in unknown environments in the spirit of comparative (as opposed to competitive) analysis, with the goal of obtaining a benchmark that is, at least conceptually, attainable. Second, by leveraging this theoretical benchmark, we derive a pseudo-optimal class of policies that can seamlessly incorporate any amount of prior or learned information while still guaranteeing the robot never collides. Finally, we demonstrate the practicality of our algorithmic approach in numerical experiments using a range of environment types and dynamics, including a comparison with a state of the art method. A key aspect of our framework is that it automatically and implicitly weighs exploration versus exploitation in a way that is optimal with respect to the information available. I. INTRODUCTIONThis paper addresses the problem of planning a safe (i.e., collision-free) trajectory from an initial state to a goal region when the obstacles in between are a priori unknown and are instead revealed online, e.g., through line-of-sight perception. A fundamental challenge is that, unlike motion planning with known obstacles or obstacles whose uncertainty is fully modeled probabilistically, when parts of the configuration space are simply unknown it is not even clear what an optimal policy to strive for (through, e.g., asymptotic convergence) is. One of the main goals of this paper is to address this shortcoming in the literature by defining a notion of optimality for use as a benchmark and also for conceptual guidance. We then follow this conceptual guidance to propose a novel algorithm for planning in unknown environments that produces low-cost solutions by flexibly incorporating side information about the environment, is guaranteed to be collision-free (even if the side information is incorrect), and requires on the order of 0.5-1s of (serial) computation time per action.Related Work: Most algorithms for motion planning assume full knowledge of the environment, and can not be used, at least directly, to find a motion plan within an incomplete map [17,18]. A common heuristic approach is to set temporary goals along the way to the goal region, and re-compute in

show abstract

Motion Trajectory Sequence-Based Map Matching Assisted Indoor Autonomous Mobile Robot Positioning

Zhang

et al. 2018

Algorithms and Architectures for Parallel Processing

View full text Add to dashboard Cite

Bayesian Learning for Safe High-Speed Navigation in Unknown Environments

Cited by 55 publications

References 23 publications

Safe Visual Navigation via Deep Learning and Novelty Detection

Safe Visual Navigation via Deep Learning and Novelty Detection

Safe Motion Planning in Unknown Environments: Optimality Benchmarks and Tractable Policies

Motion Trajectory Sequence-Based Map Matching Assisted Indoor Autonomous Mobile Robot Positioning

Contact Info

Product

Resources

About