22,122 Hits in 5.9 sec

Towards learning hierarchical skills for multi-phase manipulation tasks

Oliver Kroemer, Christian Daniel, Gerhard Neumann, Herke van Hoof, Jan Peters
2015 2015 IEEE International Conference on Robotics and Automation (ICRA)  
In this paper, we present an approach for exploiting the phase structure of tasks in order to learn manipulation skills more efficiently.  ...  Most manipulation tasks can be decomposed into a sequence of phases, where the robot's actions have different effects in each phase.  ...  CONCLUSION We proposed a method for learning hierarchical manipulation skills that exploit the phase structure of tasks.  ... 
doi:10.1109/icra.2015.7139389 dblp:conf/icra/KroemerDNH015 fatcat:3cbk6ftlxbc5bi6llcevwjfkxu

Multi-Phase Multi-Objective Dexterous Manipulation with Adaptive Hierarchical Curriculum [article]

Lingfeng Tao, Jiucai Zhang, Xiaoli Zhang
2022 arXiv   pre-print
Dexterous manipulation tasks usually have multiple objectives, and the priorities of these objectives may vary at different phases of a manipulation task.  ...  To solve this problem, we develop a novel Adaptive Hierarchical Reward Mechanism (AHRM) to guide the DRL agent to learn manipulation tasks with multiple prioritized objectives.  ...  learned skills to improve learning efficiency.  ... 
arXiv:2205.13441v2 fatcat:cqfoxm7g6ngtfoi36cd73wd66q

Towards Hierarchical Task Decomposition using Deep Reinforcement Learning for Pick and Place Subtasks [article]

Luca Marzari, Ameya Pore, Diego Dall'Alba, Gerardo Aragon-Camarasa, Alessandro Farinelli, Paolo Fiorini
2021 arXiv   pre-print
To overcome these limitations, we propose a multi-subtask reinforcement learning methodology where complex pick and place tasks can be decomposed into low-level subtasks.  ...  Deep Reinforcement Learning (DRL) is emerging as a promising approach to generate adaptive behaviors for robotic platforms.  ...  Devine et al. explored modular neural network policies to learn transferable skills for multitask and multi-robot [8] .  ... 
arXiv:2102.04022v3 fatcat:ifhtnevzsffynkned4aayxd4ay

Skill-based Meta-Reinforcement Learning [article]

Taewook Nam, Shao-Hua Sun, Karl Pertsch, Sung Ju Hwang, Joseph J Lim
2022 arXiv   pre-print
meta-learning and the usage of offline datasets, while prior approaches in RL, meta-RL, and multi-task RL require substantially more environment interactions to solve the tasks.  ...  Specifically, we propose to (1) extract reusable skills and a skill prior from offline datasets, (2) meta-train a high-level policy that learns to efficiently compose learned skills into long-horizon behaviors  ...  E.1.1 SKILL PRIOR We followed architecture and learning procedure of Pertsch et al. (2020) for learning a low-level skill policy and a skill prior.  ... 
arXiv:2204.11828v1 fatcat:rfdkkgesorb4liil44qb53kpba

Autonomous robotic valve turning: A hierarchical learning approach

Seyed Reza Ahmadzadeh, Petar Kormushev, Darwin G. Caldwell
2013 2013 IEEE International Conference on Robotics and Automation  
Autonomous valve turning is an extremely challenging task for an Autonomous Underwater Vehicle (AUV).  ...  In the first layer, the robot acquires the motor skills of approaching and grasping the valve by kinesthetic teaching.  ...  CONCLUSIONS We proposed a hierarchical learning approach to deal with the challenging task of autonomous valve turning.  ... 
doi:10.1109/icra.2013.6631235 dblp:conf/icra/AhmadzadehKC13 fatcat:mkidjok4hfgknjsvmve6ogpilu

Stochastic Neural Networks for Hierarchical Reinforcement Learning [article]

Carlos Florensa, Yan Duan, Pieter Abbeel
2017 arXiv   pre-print
To tackle these important problems, we propose a general framework that first learns useful skills in a pre-training environment, and then leverages the acquired skills for learning faster in downstream  ...  Our approach brings together some of the strengths of intrinsic motivation and hierarchical methods: the learning of useful skill is guided by a single proxy reward, the design of which requires very minimal  ...  The batch size and the maximum path length for the pre-train task are also the ones used in the benchmark : 50,000 and 500 respectively. For the downstream tasks, see Tab. 1.  ... 
arXiv:1704.03012v1 fatcat:a5no4ikvangldndvesd3blz6va

One-Shot Hierarchical Imitation Learning of Compound Visuomotor Tasks [article]

Tianhe Yu, Pieter Abbeel, Sergey Levine, Chelsea Finn
2018 arXiv   pre-print
We consider the problem of learning multi-stage vision-based tasks on a real robot from a single video of a human performing the task, while leveraging demonstration data of subtasks with other objects  ...  To address these challenges, we propose a method that learns both how to learn primitive behaviors from video demonstrations and how to dynamically compose these behaviors to perform multi-stage tasks  ...  We also acknowledge NVIDIA, Amazon, and Google for equipment and computing support.  ... 
arXiv:1810.11043v1 fatcat:55c3kfhp4zb2tpaqb4i3nnfsxa

Learning Compositional Neural Programs for Continuous Control [article]

Thomas Pierrot, Nicolas Perrin, Feryal Behbahani, Alexandre Laterre, Olivier Sigaud, Karim Beguir, Nando de Freitas
2021 arXiv   pre-print
First, we use off-policy reinforcement learning algorithms with experience replay to learn a set of atomic goal-conditioned policies, which can be easily repurposed for many tasks.  ...  We empirically show that AlphaNPI-X can effectively learn to tackle challenging sparse manipulation tasks, such as stacking multiple blocks, where powerful model-free baselines fail.  ...  Towards practical multi-object manipulation using relational reinforcement learning. arXiv preprint arXiv:1912.11032, 2019.  ... 
arXiv:2007.13363v2 fatcat:ywoyvjzfvngxbfjix4suwhxlmq

Deep Imitation Learning for Bimanual Robotic Manipulation [article]

Fan Xie, Alexander Chowdhury, M. Clara De Paolis Kaluza, Linfeng Zhao, Lawson L.S. Wong, Rose Yu
2020 arXiv   pre-print
We present a deep imitation learning framework for robotic bimanual manipulation in a continuous state-action space.  ...  A core challenge is to generalize the manipulation skills to objects in different locations.  ...  Methodology We propose a hierarchical framework for planning and control in bimanual manipulation, outlined in Figure 2 .  ... 
arXiv:2010.05134v2 fatcat:ri4rfj63mrdrhlgrrhz5riwfhy

Augmenting Reinforcement Learning with Behavior Primitives for Diverse Manipulation Tasks [article]

Soroush Nasiriany and Huihan Liu and Yuke Zhu
2022 arXiv   pre-print
While deep reinforcement learning methods have recently emerged as a promising paradigm for automating manipulation behaviors, they usually fall short in long-horizon tasks due to the exploration burden  ...  Realistic manipulation tasks require a robot to interact with an environment with a prolonged sequence of motor actions.  ...  for the contact-rich insertion phase.  ... 
arXiv:2110.03655v3 fatcat:y4pnz76ujva23eknhzrpz2wntm

Unsupervised Reinforcement Learning for Transferable Manipulation Skill Discovery [article]

Daesol Cho, Jigang Kim, H. Jin Kim
2022 arXiv   pre-print
manipulation tasks with the learned task-agnostic skills.  ...  For exploiting such benefits also in robotic manipulation, we propose an unsupervised method for transferable manipulation skill discovery that ties structured exploration toward interacting behavior and  ...  manipulation tasks with the learned task-agnostic skills.  ... 
arXiv:2204.13906v1 fatcat:lroylkcvmffotlkh6rsysqyhkm

Relay Policy Learning: Solving Long-Horizon Tasks via Imitation and Reinforcement Learning [article]

Abhishek Gupta, Vikash Kumar, Corey Lynch, Sergey Levine, Karol Hausman
2019 arXiv   pre-print
We present relay policy learning, a method for imitation and reinforcement learning that can solve multi-stage, long-horizon robotic tasks.  ...  This general and universally-applicable, two-phase approach consists of an imitation learning stage that produces goal-conditioned hierarchical policies, and a reinforcement learning phase that finetunes  ...  We also thank Robotics at Google for a wonderful research atmosphere.  ... 
arXiv:1910.11956v1 fatcat:nefhvppcgvgsnk3vuiabd32ez4

Is Curiosity All You Need? On the Utility of Emergent Behaviours from Curious Exploration [article]

Oliver Groth, Markus Wulfmeier, Giulia Vezzani, Vibhavari Dasagi, Tim Hertweck, Roland Hafner, Nicolas Heess, Martin Riedmiller
2021 arXiv   pre-print
Instead, we propose to shift the focus towards retaining the behaviours which emerge during curiosity-based learning.  ...  We argue that merely using curiosity for fast environment exploration or as a bonus reward for a specific task does not harness the full potential of this technique and misses useful skills.  ...  We would also like to thank Andrea Huber for facilitating all organisational aspects of this collaboration.  ... 
arXiv:2109.08603v1 fatcat:umza7yt35vbdvfdz3agr3v4dre

Learning Portable Symbolic Representations

Steven James
2018 Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence  
To date, however, research has investigated learning such representations for a single specific task. Our research focuses on approaches to learning these models in a domain-independent manner.  ...  We intend to use these symbolic models to build even higher levels of abstraction, creating a hierarchical representation which could be used to solve complex tasks.  ...  The first phase learns lifted symbolic rules which can be transferred between tasks, while the second combines these rules with problem-specific information to instantiate them for the current task.  ... 
doi:10.24963/ijcai.2018/826 dblp:conf/ijcai/James18 fatcat:7bme73ud4bhpna5g43dzuqm6q4

ASC me to Do Anything: Multi-task Training for Embodied AI [article]

Jiasen Lu, Jordi Salvador, Roozbeh Mottaghi, Aniruddha Kembhavi
2022 arXiv   pre-print
We propose Atomic Skill Completion (ASC), an approach for multi-task training for Embodied AI, where a set of atomic skills shared across multiple tasks are composed together to perform the tasks.  ...  In this paper, our goal is to leverage these shared skills to learn to perform multiple tasks jointly.  ...  tasks. [36] learn skills for navigation via meta-reinforcement learning.  ... 
arXiv:2202.06987v1 fatcat:2df5b5dkyzcgphfr7u263eyegq
« Previous Showing results 1 — 15 out of 22,122 results