Deep Local Trajectory Replanning and Control for Robot Navigation

Ashwini, Pokle,; Martí­n-Martí­n, Roberto; Goebel, Patrick; Chow, Vincent; Ewald, Hans M.; Yang, Jianji; Wang, Zhenkai; Sadeghian, Amir; Sadigh, Dorsa; Savarese, Silvio; Vázquez, Marynel

doi:10.1109/icra.2019.8794062

Cited by 67 publications

(44 citation statements)

References 59 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This two phase training procedure allows the user to focus on providing good demonstrations of their navigation preferences in the second phase, that will be used to modify a "vanilla" navigation policy trained in the first phase. The total number of training demonstrations required by our proposed method is far fewer than other works that utilize the classical navigation modules in their training procedure (Gao et al, 2017;Pokle et al, 2019). This is because neither one of the training procedures used in our method is trying to create a policy that completely replaces classical navigation modules.…”

Section: Training Proceduresmentioning

confidence: 95%

“…They focused on a general navigation task rather than consideration of how this setup might be used as a mechanism for Imitation Learning. Gao et al's intention-net (Gao et al, 2017), and Pokle et al's work (Pokle et al, 2019) are similar to our own approach as they utilize the global planner provide the general direction that a robot should travel to reach a desired destination in a known environment. These methods attempt to address the shortcomings of classical navigation with Machine Learning at the cost of high training time, complexity, and data efficiency by training models to replace the functionality of proven low level controllers used in classical navigation techniques.…”

Section: Related Workmentioning

confidence: 98%

See 1 more Smart Citation

Improving Autonomous Robotic Navigation Using Imitation Learning

et al. 2021

View full text Add to dashboard Cite

Autonomous navigation to a specified waypoint is traditionally accomplished with a layered stack of global path planning and local motion planning modules that generate feasible and obstacle-free trajectories. While these modules can be modified to meet task-specific constraints and user preferences, current modification procedures require substantial effort on the part of an expert roboticist with a great deal of technical training. In this paper, we simplify this process by inserting a Machine Learning module between the global path planning and local motion planning modules of an off-the shelf navigation stack. This model can be trained with human demonstrations of the preferred navigation behavior, using a training procedure based on Behavioral Cloning, allowing for an intuitive modification of the navigation policy by non-technical users to suit task-specific constraints. We find that our approach can successfully adapt a robot’s navigation behavior to become more like that of a demonstrator. Moreover, for a fixed amount of demonstration data, we find that the proposed technique compares favorably to recent baselines with respect to both navigation success rate and trajectory similarity to the demonstrator.

show abstract

Section: Training Proceduresmentioning

confidence: 95%

Section: Related Workmentioning

confidence: 98%

Improving Autonomous Robotic Navigation Using Imitation Learning

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Besides of classic algorithms, some systems involve machine learning methods or neural networks. So, in [16] it is proposed to leverage neural networks to model situations with people, occurring in the operational area of the robot. To do this, authors propose a global planner, based on the Dijkstra's algorithm to find the shortest path.…”

Section: Related Workmentioning

confidence: 99%

Algorithm of Target Point Assignment for Robot Path Planning Based on Costmap Data

Kuznetsov

Kozyr

Levonevskiy

2021

Smart Innovation, Systems and Technologies

View full text Add to dashboard Cite

Most mobile robots which use the operating systems (ROS) employ costmap-based path planner for navigation in environment. Besides the systems, in which robotic devices function in completely autonomous manner, also such systems exist, where user specifies in a self-contained interface the target point for robot motion, without using any costmap-related data. In such settings, the robot sometimes cannot reach the user-specified point because it is treated as an obstacle on the costmap. To solve this problem an algorithm was developed, which accepts the coordinates of the target point, specified through user interface, and finds possible target points in vicinity of the specified one, regarding the predefined distance constraints. Conceptually, the search of possible points consists in looping through costmap cells of the global cost planner in orthogonally related directions from the obtained point and in defining of the preferable target points, upon reaching which the task is considered to be completed. In this paper, the algorithm was tested as part of the navigational module of the robot. Experiments were performed, particularly, in the Gazebo simulation environment; robot model TURTLEBOT3 Waffle Pi was also utilized. The experimental results showed that this module can be combined with any planner. When the value of validation step was set at the minimum value of 0.01 m, it was required no more than 55 ms to find a target point. The minimum duration of such search was 1 ms.

show abstract

“…The authors use depth data to train the network and the social force model to generate a large set of training data. Pokle et al [9] designed a local controller to determine the robot's velocity commands and predicting a local motion plan, while considering the trajectories of surrounding humans. These supervised learning methods all depend on the teacher, e.g., controls provided by humans, a global path planner, or a well-tuned optimization, while the goal of our work is to enable the robot to learn by itself while navigating in the environment.…”

Section: Related Workmentioning

confidence: 99%

Deep Reinforcement Learning for Navigation in Cluttered Environments

Seguchi¹,

Gesing²,

Bennewitz³

et al. 2020

Computer Science &Amp; Information Technology (CS &Amp; IT)

View full text Add to dashboard Cite

Collision-free motion is essential for mobile robots. Most approaches to collision-free and efficient navigation with wheeled robots require parameter tuning by experts to obtain good navigation behavior. In this paper, we aim at learning an optimal navigation policy by deep reinforcement learning to overcome this manual parameter tuning. Our approach uses proximal policy optimization to train the policy and achieve collision-free and goal-directed behavior. The output of the learned network are the robot's translational and angular velocities for the next time step. Our method combines path planning on a 2D grid with reinforcement learning and does not need any supervision. Our network is first trained in a simple environment and then transferred to scenarios of increasing complexity. We implemented our approach in C++ and Python for the Robot Operating System (ROS) and thoroughly tested it in several simulated as well as real-world experiments. The experiments illustrate that our trained policy can be applied to solve complex navigation tasks. Furthermore, we compare the performance of our learned controller to the popular dynamic window approach (DWA) of ROS. As the experimental results show, a robot controlled by our learned policy reaches the goal significantly faster compared to using the DWA by closely bypassing obstacles and thus saving time.

show abstract

Deep Local Trajectory Replanning and Control for Robot Navigation

Cited by 67 publications

References 59 publications

Improving Autonomous Robotic Navigation Using Imitation Learning

Improving Autonomous Robotic Navigation Using Imitation Learning

Algorithm of Target Point Assignment for Robot Path Planning Based on Costmap Data

Deep Reinforcement Learning for Navigation in Cluttered Environments

Contact Info

Product

Resources

About