Universal Planning Networks

Srinivas, Aravind; Jabri, Allan; Abbeel, Pieter; Levine, Sergey; Finn, Chelsea

doi:10.48550/arxiv.1804.00645

Cited by 34 publications

(62 citation statements)

References 35 publications

(49 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Additionally, we compare to two approaches that plan over a finite horizon by gradient decent. The first approach is based on the Universal Planning Network with a horizon of 5 (UPN) [11]. We also compare to an approach that extends UPN with neuromodulation for the integrated policy and dynamics model along with task embedding (TE-CPN).…”

Section: A Methods For Comparisonmentioning

confidence: 99%

“…The idea can be extended by using a dynamics model to predict the latent state resulting from an action. Prior work [11], [15], [16], [17], [18] has demonstrated the value of using the dynamics model to unroll the policy over a planning horizon with the goal of minimizing the distance between the latent representations of the final predicted state and the goal.…”

Section: Planning By Backpropagationmentioning

confidence: 99%

“…The goal of this meta-learning approach is to directly optimize for plannable representations [11]. The approach is inspired by other well known approaches like MAML [7] that optimize a model for an ability to quickly fine tune for a new task.…”

Section: Universal Planning Networkmentioning

confidence: 99%

“…To solve this problem, We introduce an approach called contextual planning networks (CPN) that learns a combined representation of the policy and dynamics using an objective that directly optimizes for the ability to plan towards a goal state represented as an image. Our approach draws inspiration from several previous approaches [11], [6], [8] and builds upon them. The approach combines planning by backpropagation, with task embedding, and neuromodulation.…”

Section: Introductionmentioning

confidence: 99%

“…Using an image as a task goal has unique challenges as many pixel level details of the image may be irrelevant or misleading for the task. Instead, many approaches have explored latent representations of the images that are intended to capture the salient aspects of the task from the image [13], [11]. To discover these latent representations, prior work has mostly focused on unsupervised objectives or objectives that are disconnected from learning the policy [13], [14].…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Visual Goal-Directed Meta-Learning with Contextual Planning Networks

Rivera¹,

Handelman²

2021

Preprint

View full text Add to dashboard Cite

The goal of meta-learning is to generalize to new tasks and goals as quickly as possible. Ideally, we would like approaches that generalize to new goals and tasks on the first attempt. Toward that end, we introduce contextual planning networks (CPN). Tasks are represented as goal images and used to condition the approach. We evaluate CPN along with several other approaches adapted for zero-shot goal-directed meta-learning. We evaluate these approaches across 24 distinct manipulation tasks using Metaworld benchmark tasks. We found that CPN outperformed several approaches and baselines on one task and was competitive with existing approaches on others. We demonstrate the approach on a physical platform on Jenga tasks using a Kinova Jaco robotic arm.

show abstract

Section: A Methods For Comparisonmentioning

confidence: 99%

Section: Planning By Backpropagationmentioning

confidence: 99%

Section: Universal Planning Networkmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Visual Goal-Directed Meta-Learning with Contextual Planning Networks

Rivera¹,

Handelman²

2021

Preprint

View full text Add to dashboard Cite

show abstract

Towards Learning Abstract Representations for Locomotion Planning in High-dimensional State Spaces

Klamt

Behnke

2019

2019 International Conference on Robotics and Automation (ICRA)

View full text Add to dashboard Cite

Ground robots which are able to navigate a variety of terrains are needed in many domains. One of the key aspects is the capability to adapt to the ground structure, which can be realized through movable body parts coming along with additional degrees of freedom (DoF). However, planning respective locomotion is challenging since suitable representations result in large state spaces. Employing an additional abstract representation-which is coarser, lower-dimensional, and semantically enriched-can support the planning.While a desired robot representation and action set of such an abstract representation can be easily defined, the cost function requires large tuning efforts. We propose a method to represent the cost function as a CNN. Training of the network is done on generated artificial data, while it generalizes well to the abstraction of real world scenes. We further apply our method to the problem of search-based planning of hybrid drivingstepping locomotion. The abstract representation is used as a powerful informed heuristic which accelerates planning by multiple orders of magnitude. Abstract representation HeuristicPlanner (e.g., A*, RRT, PRM) Path

show abstract

A Data-Efficient Framework for Training and Sim-to-Real Transfer of Navigation Policies

Bharadhwaj

Wang

Bengio

et al. 2019

2019 International Conference on Robotics and Automation (ICRA)

View full text Add to dashboard Cite

Learning effective visuomotor policies for robots purely from data is challenging, but also appealing since a learning-based system should not require manual tuning or calibration. In the case of a robot operating in a real environment the training process can be costly, time-consuming, and even dangerous since failures are common at the start of training. For this reason, it is desirable to be able to leverage simulation and off-policy data to the extent possible to train the robot. In this work, we introduce a robust framework that plans in simulation and transfers well to the real environment. Our model incorporates a gradient-descent based planning module, which, given the initial image and goal image, encodes the images to a lower dimensional latent state and plans a trajectory to reach the goal. The model, consisting of the encoder and planner modules, is trained through a meta-learning strategy in simulation first. We subsequently perform adversarial domain transfer on the encoder by using a bank of unlabelled but random images from the simulation and real environments to enable the encoder to map images from the real and simulated environments to a similarly distributed latent representation. By fine tuning the entire model (encoder + planner) with far fewer real world expert demonstrations, we show successful planning performances in different navigation tasks.

show abstract

Universal Planning Networks

Cited by 34 publications

References 35 publications

Visual Goal-Directed Meta-Learning with Contextual Planning Networks

Visual Goal-Directed Meta-Learning with Contextual Planning Networks

Towards Learning Abstract Representations for Locomotion Planning in High-dimensional State Spaces

A Data-Efficient Framework for Training and Sim-to-Real Transfer of Navigation Policies

Contact Info

Product

Resources

About