Robot Motion Planning in Learned Latent Spaces

@article{Ichter2019RobotMP,
  title={Robot Motion Planning in Learned Latent Spaces},
  author={Brian Ichter and Marco Pavone},
  journal={IEEE Robotics and Automation Letters},
  year={2019},
  volume={4},
  pages={2407-2414}
}
This letter presents latent sampling-based motion planning (L-SBMP), a methodology toward computing motion plans for complex robotic systems by learning a plannable latent representation. Recent works in control of robotic systems have effectively leveraged local, low-dimensional embeddings of high-dimensional dynamics. In this letter, we combine these recent advances with techniques from sampling-based motion planning (SBMP) in order to design a methodology capable of planning for high… 

Figures from this paper

Learning an Optimal Sampling Distribution for Efficient Motion Planning
TLDR
A learning-based approach with policy improvement to compute an optimal sampling distribution for use in sampling-based motion planners, motivated by the challenge of whole-body planning for a 31 degree-of-freedom mobile robot.
Reaching Through Latent Space: From Joint Statistics to Path Planning in Manipulation
TLDR
A novel approach to path planning for robotic manipulators is presented, in which paths are produced via iterative optimisation in the latent space of a generative model of robot poses, which leverages gradients through learned models that provide a simple way to combine goal reaching objectives with constraint satisfaction, even in the presence of otherwise non-differentiable constraints.
Motion Planning Networks: Bridging the Gap Between Learning-Based and Classical Motion Planners
TLDR
This article describes motion planning networks (MPNet), a computationally efficient, learning-based neural planner for solving motion planning problems, and shows that worst-case theoretical guarantees can be proven if this neural network strategy is merged with classical sample-based planners in a hybrid approach.
Learning Equality Constraints for Motion Planning on Manifolds
TLDR
This work considers the problem of learning representations of constraints from demonstrations with a deep neural network, which it calls Equality Constraint Manifold Neural Network (ECoMaNN), to learn a level-set function of the constraint suitable for integration into a constrained sampling-based motion planner.
MPC-MPNet: Model-Predictive Motion Planning Networks for Fast, Near-Optimal Planning Under Kinodynamic Constraints
TLDR
This work presents a scalable, imitation learning-based, Model-Predictive Motion Planning Networks framework that quickly finds near-optimal path solutions with worst-case theoretical guarantees under kinodynamic constraints for practical underactuated systems.
Harnessing Reinforcement Learning for Neural Motion Planning
TLDR
This work proposes a modification of the popular DDPG RL algorithm that is tailored to motion planning domains, by exploiting the known model in the problem and the set of solved plans in the data, and shows that the algorithm can plan significantly faster on novel domains than off-the-shelf sampling based motion planners.
A review of mobile robot motion planning methods: from classical motion planning workflows to reinforcement learning-based architectures
TLDR
This paper summarizes the representative and state-of-the-art works for each submodule of the classical motion planning architecture and analyzes their performance features, focusing on summarizing RLbased motion planning approaches, including motion planners combined with RL improvements, map-free RL-based motion planners, and multi-robot cooperative planning methods.
Enabling Visual Action Planning for Object Manipulation through Latent Space Roadmap
TLDR
A thorough investigation is presented of the framework for visual action plan- ning of complex manipulation tasks with high-dimensional state spaces, focusing on manipulation of deformable objects on simulated box stacking and rope/box manipulation tasks, and a folding task executed on a real robot.
Learning to Predict Ego-Vehicle Poses for Sampling-Based Nonholonomic Motion Planning
TLDR
A data-driven approach utilizing a deep convolutional neural network allowing to guide the motion planner efficiently toward the optimal solution and reduces the computation time by up to an order of magnitude and yields a faster convergence to a lower cost.
Deep Visual Reasoning: Learning to Predict Action Sequences for Task and Motion Planning from an Initial Scene Image
TLDR
A deep convolutional recurrent neural network that predicts action sequences for task and motion planning (TAMP) from an initial scene image such that ideally only one motion planning problem has to be solved to find a solution to the overall TAMP problem.
...
...

References

SHOWING 1-10 OF 26 REFERENCES
Learning Sampling Distributions for Robot Motion Planning
TLDR
This paper proposes a methodology for nonuniform sampling, whereby a sampling distribution is learned from demonstrations, and then used to bias sampling, resulting in an order of magnitude improvement in terms of success rate and convergence to the optimal cost.
Universal Planning Networks
TLDR
This work finds that the representations learned are not only effective for goal-directed visual imitation via gradient-based trajectory optimization, but can also provide a metric for specifying goals using images.
Deep visual foresight for planning robot motion
  • Chelsea FinnS. Levine
  • Computer Science
    2017 IEEE International Conference on Robotics and Automation (ICRA)
  • 2017
TLDR
This work develops a method for combining deep action-conditioned video prediction models with model-predictive control that uses entirely unlabeled training data and enables a real robot to perform nonprehensile manipulation — pushing objects — and can handle novel objects not seen during training.
Motion Planning Networks
TLDR
This work presents Motion Planning Networks (MPNet), a neural network-based novel planning algorithm that encodes the given workspaces directly from a point cloud measurement and generates the end-to-end collision-free paths for the given start and goal configurations.
Multimodal Probabilistic Model-Based Planning for Human-Robot Interaction
TLDR
The approach is to learn multimodal probability distributions over future human actions from a dataset of human-human exemplars and perform real-time robot policy construction in the resulting environment model through massively parallel sampling of human responses to candidate robot action sequences.
Fastron : A Learning-Based Configuration Space Model for Rapid Collision Detection for Gross Motion Planning in Changing Environments
Collision detection is a necessary but costly step for sampling-based motion planners, such as Rapidly-Exploring Random Trees [7]. Motion planning is typically performed in configuration space
Deep spatial autoencoders for visuomotor learning
TLDR
This work presents an approach that automates state-space construction by learning a state representation directly from camera images by using a deep spatial autoencoder to acquire a set of feature points that describe the environment for the current task, such as the positions of objects.
Learning visual representations for perception-action systems
TLDR
This work argues in favor of task-specific, learn-able representations for vision as a sensory modality for systems that interact flexibly with uncontrolled environments and develops a grasp density for object detection in a novel scene.
Randomized kinodynamic planning
  • S. LaValleJ. Kuffner
  • Mathematics
    Proceedings 1999 IEEE International Conference on Robotics and Automation (Cat. No.99CH36288C)
  • 1999
TLDR
A state-space perspective on the kinodynamic planning problem is presented, and a randomized path planning technique that computes collision-free kinodynamic trajectories for high degree-of-freedom problems is introduced.
Self-Consistent Trajectory Autoencoder: Hierarchical Reinforcement Learning with Trajectory Embeddings
TLDR
This work shows that it can learn continuous latent representations of trajectories, which are effective in solving temporally extended and multi-stage problems and provides a built-in prediction mechanism, by predicting the outcome of closed loop policy behavior.
...
...