Multi-Object Rearrangement with Monte Carlo Tree Search: A Case Study on Planar Nonprehensile Sorting

Song, Haoran; Haustein, Joshua A.; Yuan, Weihao; Hang, Kaiyu; Wang, Michael Yu; Kragić, Danica; Stork, Johannes A.

doi:10.1109/iros45743.2020.9341532

Cited by 42 publications

(33 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Obtaining Skill Effects. Many prior works used simulated skill outcomes during planning [19], [20], [21], [14]. This can be prohibitively expensive to perform online, depending on the complexity of simulation and the duration of each skill.…”

Section: Related Workmentioning

confidence: 99%

Search-Based Task Planning with Learned Skill Effect Models for Lifelong Robotic Manipulation

Liang¹,

Sharma²,

Alex³

et al. 2021

Preprint

View full text Add to dashboard Cite

Lifelong-learning robots need to be able to acquire new skills and plan for new tasks over time. Prior works on planning with skills often make assumptions on the structure of skills and tasks, like subgoal skills, shared skill implementations, or learning task-specific plan skeletons, that limit their application to new and different skills and tasks. By contrast, we propose doing task planning by jointly searching in the space of skills and their parameters with skill effect models learned in simulation. Our approach is flexible about skill parameterizations and task specifications, and we use an iterative training procedure to efficiently generate relevant data to train such models. Experiments demonstrate the ability of our planner to integrate new skills in a lifelong manner, finding new task strategies with lower costs in both train and test tasks. We additionally show that our method can transfer to the real world without further fine-tuning.

show abstract

Section: Related Workmentioning

confidence: 99%

Search-Based Task Planning with Learned Skill Effect Models for Lifelong Robotic Manipulation

Liang¹,

Sharma²,

Alex³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…This combination is dominated by tree search planners to guide RL policy search [3,53], mostly by the adoption of Monte Carlo Tree Search (MCTS) that uses Upper Confidence Bounds (UCB) to balance between state space exploration and rewards exploitation [27,48]. Song et al [54] apply this concept to physics-based manipulation domains for solving rearrangement tasks that necessitate a long sequence of actions. Albeit the use of neural networks that are recursively trained over previous iterations of the generated plans to speed up the search [8,32], the computation cost associated with the transition function of physics models remains prohibitively expensive for this process to run in closed-loop in real-time.…”

Section: Related Workmentioning

confidence: 99%

Learning image-based Receding Horizon Planning for manipulation in clutter

Bejjani

Leonetti

Dogar

2021

Robotics and Autonomous Systems

View full text Add to dashboard Cite

“…Minimizing the number of buffers can enhance the efficiency of solving non-monotone problems [23]. Recently Monte Carlo Tree Search (MCTS) [24]- [26] and deep learning [27], [28] have been applied on rearrangement. The above methods may not be always transferable to the harder, confined setup considered here.…”

Section: Related Workmentioning

confidence: 99%

Efficient and High-quality Prehensile Rearrangement in Cluttered and Confined Spaces

Wang¹,

Miao²,

Bekris³

2021

Preprint

View full text Add to dashboard Cite

Prehensile object rearrangement in cluttered and confined spaces has broad applications but is also challenging. For instance, rearranging products in a grocery or home shelf means that the robot cannot directly access all objects and has limited free space. This is harder than tabletop rearrangement where objects are easily accessible with top-down grasps, which simplifies robot-object interactions. This work focuses on problems where such interactions are critical for completing tasks and extends state-of-the-art results in rearrangement planning. It proposes a new efficient and complete solver under general constraints for monotone instances, which can be solved by moving each object at most once. The monotone solver reasons about robot-object constraints and uses them to effectively prune the search space. The new monotone solver is integrated with a global planner to solve non-monotone instances with high-quality solutions fast. Furthermore, this work contributes an effective pre-processing tool to speed up arm motion planning for rearrangement in confined spaces. The pre-processing tool provide significant speed-ups (49.1% faster on average) in online query resolution. Comparisons in simulations further demonstrate that the proposed monotone solver, equipped with the pre-processing tool, results in 57.3% faster computation and 3 times higher success rate than alternatives. Similarly, the resulting global planner is computationally more efficient and has a higher success rate given the more powerful monotone solver and the pre-processing tool, while producing high-quality solutions for non-monotone instances (i.e., only 1.3 buffers are needed on average). Videos of demonstrating solutions on a real robotic system and codes can be found at https://github.com/Rui1223/uniform object rearrangement.

show abstract

Multi-Object Rearrangement with Monte Carlo Tree Search: A Case Study on Planar Nonprehensile Sorting

Cited by 42 publications

References 28 publications

Search-Based Task Planning with Learned Skill Effect Models for Lifelong Robotic Manipulation

Search-Based Task Planning with Learned Skill Effect Models for Lifelong Robotic Manipulation

Learning image-based Receding Horizon Planning for manipulation in clutter

Efficient and High-quality Prehensile Rearrangement in Cluttered and Confined Spaces

Contact Info

Product

Resources

About