Fast simulation of vehicles with non-deformable tracks

Abstract

This paper presents a novel technique that allows for both computationally fast and sufficiently plausible simulation of vehicles with non-deformable tracks. The method is based on an effect we have called Contact Surface Motion. A comparison with several other methoddds for simulation of tracked vehicle dynamics is presented with the aim to evaluate methods that are available off-the-shelf or with minimum effort in general-purpose robotics simulators. The proposed method is implemented as a plugin for the open-source physics-based simulator Gazebo using the Open Dynamics Engine.

Staircase Traversal via Reinforcement Learning for Active Reconfiguration of Assistive Robots

“…Among various alternatives for imitating mobile robot tracks, we employ the common approximation of representing tracks by an ensemble of overlapping, non self-colliding, tracked wheels (cf. [6]).…”

“…Broadly speaking, this complicates portability to platforms with different or unknown kinematics or poor environment observability, weakly taking into consideration the dynamics of physical interaction. On the other hand, learning-based control of flippers [5] allows to make less restrictive hypotheses with respect to variation in the environment or the robotic kinematics and focuses more on task constraints such as safety [6]. A more recent approach [7] for navigation in multi floor environment, that has shown good performance on more than hundreds of staircase instances, relies on the passive adaptation to obstacles which cannot be generalized to assistive robots due to high movement stochasticity.…”

Data-Driven Policy Transfer With Imprecise Perception Simulation

“…The aim of this work is to learn motion control policy for four independently articulated flippers of a tracked skidsteering robot shown in Figure 2. The proposed method exploits an analytically non-differentiable dynamics-enginebased simulator of the real platform [3]. The learned policy maps the local height map and pose of the robot to desired motion of the flippers, which assures smooth traversal over complex unstructured terrain.…”

“…The complexity of track-terrain interactions [3] slows the simulation speed down to real-time, therefore collecting a huge number of samples needed for accurate learning is impossible. Consequently, we propose coarse-to-fine policy learning, where the coarse motion planning is alternated with imitation learning and policy transfer to the real robot.…”

“…SegMatch algorithm [12] is combined with high-precision track odometry in a multi-modal fusion pipeline [13]. For simulation, we use our custom tracked vehicle dynamics model implemented in the Gazebo simulator [3]. We showed that although it is a good approximation of motion of the tracked vehicle, it is not a perfect simulation of the real world.…”