Deep Visual MPC-Policy Learning for Navigation

Abstract

Humans can routinely follow a trajectory defined by a list of images/landmarks. However, traditional robot navigation methods require accurate mapping of the environment, localization, and planning. Moreover, these methods are sensitive to subtle changes in the environment. In this paper, we propose a Deep Visual MPC-policy learning method that can perform visual navigation while avoiding collisions with unseen objects on the navigation path. Our model PoliNet takes in as input a visual trajectory and the image of the robot's current view and outputs velocity commands for a planning horizon of N steps that optimally balance between trajectory following and obstacle avoidance. PoliNet is trained using a strong image predictive model and traversability estimation model in a MPC setup, with minimal human supervision. Different from prior work, PoliNet can be applied to new scenes without retraining. We show experimentally that the robot can follow a visual trajectory when varying start position and in the presence of previously unseen obstacles. We validated our algorithm with tests both in a realistic simulation environment and in the real world. We also show that we can generate visual trajectories in simulation and execute the corresponding path in the real environment. Our approach outperforms classical approaches as well as previous learning-based baselines in success rate of goal reaching, sub-goal coverage rate, and computational load.

Interactive Gibson Benchmark: A Benchmark for Interactive Navigation in Cluttered Environments

“…The main advantage of Gibson V1 is that it generates photo-realistic virtual images for the agent. This enabled seamless sim2real transfer [14], [38]. However, Gibson V1 cannot be used as a test bed for Interactive Navigation because neither the rendering nor the assets (hundreds of 3D photo-realistic models reconstructed from real-world environments) allow for changes in the state of the environment.…”

“…These environments have the desired photo and layout realism, and provide sufficient scene complexity. They have enabled the development and benchmarking of learning-based navigation algorithms, and some have allowed relatively easy deployment of such algorithms on real robots [13], [14]. Most of these established simulators, however, fall short of providing interactivity -scans of real worlds are static, and objects cannot be manipulated.…”

Effects of a Social Force Model Reward in Robot Navigation Based on Deep Reinforcement Learning

“…Several works [4,5,6,17] for navigation have been developed using these methods to learn policies. Sometimes, these methods combine visual information to improve predictions like PoliNet [15]. Another method [9] combines Probabilistic Roadmaps and a RL based local planner to guide the robot for long-range indoor navigation.…”

OmniFlowNet: a Perspective Neural Network Adaptation for Optical Flow Estimation in Omnidirectional Images

“…Thanks to their wide field of view, they can capture their entire environment in a single image. These sensors are therefore used in many mobile robotics tasks, such as image-based navigation [1], monitoring [2] or simultaneous visual localization and mapping (SLAM) [3].…”