Optimal Robot Motion Planning in Constrained Workspaces Using Reinforcement Learning

Rousseas, Panagiotis; Bechlioulis, Charalampos P.; Kyriakopoulos, Kostas J.

doi:10.1109/iros45743.2020.9341148

Cited by 12 publications

(10 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our method’s cost function is presented again for completeness. It is clear that, although the method by Rousseas et al (2020) provides descent results, significantly better than the constant weights case, the herein proposed method exhibits a significant overall improvement.…”

Section: Resultsmentioning

confidence: 83%

“…Our method outperforms both previous ones. While it is not self-evident due to the depicted range, the proposed method assumes a maximal value of

16

, whereas the method in Rousseas et al (2020) assumes a maximal value of

20

, with similar relative cost-value distributions over the workspace.…”

Section: Resultsmentioning

confidence: 99%

“…The results for the reactive, AHPF methods are presented in Figure 13 . In the left-most figure, the method by Rousseas et al (2020) was implemented, where the weights of an AHPF assume a state-feedback form. In the central figure, we present a custom method where a constrained-optimization RL strategy is employed to extract the optimal constant AHPF weights.…”

Section: Resultsmentioning

confidence: 99%

“… Comparative Final Cost functions between three reactive methods, namely ( Rousseas et al, 2020 ) in logarithmic scale, where the AHPF weights are non-constant functions of the position (first subfigure), a custom, in-house RL scheme which computes the optimal (constant) AHPF weights (second subfigure), and the proposed method respectively (third subfigure). Our method outperforms both previous ones.…”

Section: Resultsmentioning

confidence: 99%

“…Nonetheless, note that, even though path length is a valid metric for these methods, the cost function (2) can not be directly evaluated, as they only produce feasible paths and not inputs. In previous works ( Rousseas et al, 2020 ; Rousseas et al, 2021 ; Rousseas et al, 2022 ), we have employed RRT ⋆ planners that include input-space sampling. However, due to sampling over a 4D space, the RRT ⋆ results presented therein do not consist of fair evaluations against the current method.…”

Section: Resultsmentioning

confidence: 99%

See 4 more Smart Citations

Reactive optimal motion planning for a class of holonomic planar agents using reinforcement learning with provable guarantees

Rousseas,

Bechlioulis,

Kyriakopoulos

2024

Front. Robot. AI

View full text Add to dashboard Cite

In control theory, reactive methods have been widely celebrated owing to their success in providing robust, provably convergent solutions to control problems. Even though such methods have long been formulated for motion planning, optimality has largely been left untreated through reactive means, with the community focusing on discrete/graph-based solutions. Although the latter exhibit certain advantages (completeness, complicated state-spaces), the recent rise in Reinforcement Learning (RL), provides novel ways to address the limitations of reactive methods. The goal of this paper is to treat the reactive optimal motion planning problem through an RL framework. A policy iteration RL scheme is formulated in a consistent manner with the control-theoretic results, thus utilizing the advantages of each approach in a complementary way; RL is employed to construct the optimal input without necessitating the solution of a hard, non-linear partial differential equation. Conversely, safety, convergence and policy improvement are guaranteed through control theoretic arguments. The proposed method is validated in simulated synthetic workspaces, and compared against reactive methods as well as a PRM and an RRT⋆ approach. The proposed method outperforms or closely matches the latter methods, indicating the near global optimality of the former, while providing a solution for planning from anywhere within the workspace to the goal position.

show abstract

Section: Resultsmentioning

confidence: 83%

“…Our method outperforms both previous ones. While it is not self-evident due to the depicted range, the proposed method assumes a maximal value of

16

, whereas the method in Rousseas et al (2020) assumes a maximal value of

20

, with similar relative cost-value distributions over the workspace.…”

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

See 3 more Smart Citations

Reactive optimal motion planning for a class of holonomic planar agents using reinforcement learning with provable guarantees

Rousseas,

Bechlioulis,

Kyriakopoulos

2024

Front. Robot. AI

View full text Add to dashboard Cite

show abstract

LAC-RRT: Constrained Rapidly-Exploring Random Tree With Configuration Transfer Models for Motion Planning

Ho,

King

2023

IEEE Access

View full text Add to dashboard Cite

Motion planning is challenging for robotic manipulators. Achieving high efficiency and generalization while considering various constraints simultaneously can be difficult. One common solution is to use iterative projection techniques to acquire feasible manipulator configurations and then search for a valid path to reach the goal state based on those configurations. However, such iterative techniques tend to be computationally expensive and time-consuming. The problem becomes more serious if equality constraints are involved, due to their narrower solution space. In this paper, we propose the Learning-Assisted Constrained Rapidly-Exploring Random Tree (LAC-RRT) algorithm, which employs self-supervised learning to train a model that can directly convert any sampled configuration to a new and valid configuration using feature values constrained by the imposed equality constraints, avoiding the need for iterative optimizations. Unlike other learning-based motion planning techniques, which typically solve the problem by building the constraint manifold based on a fixed set of constraints, LAC-RRT permits better generalization by allowing the equality constraints to be specified at run time. The experimental results show that the proposed LAC-RRT surpasses other approaches in most cases. Specifically, LAC-RRT can significantly reduce computation time by 80-90% for acquiring valid configurations and performing motion planning.INDEX TERMS Constrained motion planning, self-supervised learning, constraint manifold

show abstract

A Continuous Off-Policy Reinforcement Learning Scheme for Optimal Motion Planning in Simply-Connected Workspaces

Rousseas

Bechlioulis

Kyriakopoulos

2023

2023 IEEE International Conference on Robotics and Automation (ICRA)

View full text Add to dashboard Cite

Optimal Robot Motion Planning in Constrained Workspaces Using Reinforcement Learning

Cited by 12 publications

References 16 publications

Reactive optimal motion planning for a class of holonomic planar agents using reinforcement learning with provable guarantees

Reactive optimal motion planning for a class of holonomic planar agents using reinforcement learning with provable guarantees

LAC-RRT: Constrained Rapidly-Exploring Random Tree With Configuration Transfer Models for Motion Planning

A Continuous Off-Policy Reinforcement Learning Scheme for Optimal Motion Planning in Simply-Connected Workspaces

Contact Info

Product

Resources

About