Cognitive Mapping and Planning for Visual Navigation

By Saurabh Gupta, Varun Tolani, James Davidson, Sergey Levine, Rahul Sukthankar, Jitendra Malik. 2017

770 citations recorded by the source corpus.

We introduce a neural architecture for navigation in novel environments. Our proposed architecture learns to map from first-person views and plans a sequence of actions towards goals in the environment. The Cognitive Mapper and Planner (CMP) is based on two key ideas: (a) a unified joint architecture for mapping and planning, such that the mapping is driven by the needs of the task, and (b) a spatial memory with the ability to plan given an incomplete set of observations about the world. CMP constructs a top-down belief map of the world and applies a differentiable neural net planner to produce the next action at each time step. The accumulated belief of the world enables the agent to track visited regions of the environment. We train and test CMP on navigation problems in simulation environments derived from scans of real world buildings. Our experiments demonstrate that CMP outperforms alternate learning-based architectures, as well as, classical mapping and path planning approaches in many cases. Furthermore, it naturally extends to semantically specified goals, such as “going to a chair”. We also deploy CMP on physical robots in indoor environments, where it achieves reasonable performance, even though it is trained entirely in simulation.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition, Robotics & Control

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Computer Vision and Pattern Recognition

In Robotics & Control

Citations in the corpus

Vision-and-Language Navigation: Interpreting Visually-Grounded Navigation Instructions in Real Environments (2017, background): “RL in navigation A number of recent papers use reinforcement learning (RL) to train navigational agents [31, 50, 53, 62, 21], although these works do not address language instruction.”
Embodied Question Answering (2017, background): “Visual navigation typically speciﬁes agent goals either implicitly via the reward function [18, 19] (thus training a separate policy for each goal/reward), or explicitly by conditioning on goal state representations [23] like images of target objects [17, 24].”
Shaping Belief States with Generative Environment Models for RL (2019, background): “Extracting a consistent high-level representation of the environment such as the bird’s eye view "map" from merely first-person views and actions in a completely unsupervised manner is a notoriously difficult task [12] and a lot of the success in addressing this problem is due to the injection of a ”
Continuous Scene Representations for Embodied AI (2022, background): “Neural graphbased [9, 10, 43, 58] and dense representations [8, 25] have been proposed to construct the map of the environment.”
A Survey of Embodied AI: From Simulators to Research Tasks (2021, background): “The internal model can be in forms like a topological graph map [51], semantic map [52], occupancy map [53] or spatial memory [54, 55].”
Deep Drone Racing: Learning Agile Flight in Dynamic Environments (2018, background): “While being independent of any global map and position estimate, these methods are not directly applicable to our specific problem due to their high computational complexity [27, 11], their low maximum speed [13] or the inherent difficulties of generalizing to 3D motions [25, 4, 24].”
Self-Supervised Deep Reinforcement Learning with Generalized Computation Graphs for Robot Navigation (2017, background): “Model-free methods have empirically demonstrated state-ofthe-art performance in many complex tasks [6], [7], including navigation [26].”
The AdobeIndoorNav Dataset: Towards Deep Reinforcement Learning based Real-world Indoor Robot Visual Navigation (2018, background): “When navigating in an environment without a map, the robot can either incrementally build a map while navigating around [21] or simply following a reactive setting to navigate based on current state and target state [1].”
IQA: Visual Question Answering in Interactive Environments (2017, background): “[16] proposed a joint architecture for a mapper that produces a spatial memory and a planner that can plan paths.”
Habitat: A Platform for Embodied AI Research (2019, methodology): “The availability of large-scale 3D scene datasets [5, 24, 8] and community interest in active vision tasks led to a recent surge of work that resulted in the development of a variety of simulation platforms for indoor environments [17, 7, 13, 22, 26, 3, 27, 28, 21].”
Building Generalizable Agents with a Realistic and Rich 3D Environment (2018, background): “Gupta et al. (2017) learn to navigate via mapping and planning using shortest path supervision.”
MINOS: Multimodal Indoor Simulator for Navigation in Complex Environments (2017, methodology): “Note that since we consider agents acting in the continuous state space, we do not include the method of Gupta et al. (2017), which assumes a discrete gridworld-like environment.”
Stabilizing Transformers for Reinforcement Learning (2019, methodology): “Priors are induced through the design of the speciﬁc read/write operations, such as those resembling a digital computer (Wayne et al., 2018; Graves et al., 2016) or an environment map (Parisotto & Salakhutdinov, 2018; Gupta et al., 2017).”
On Evaluation of Embodied Navigation Agents (2018, background): “Agents are asked to ﬁnd their way around novel, previously unseen environments [12, 28].”
Semi-parametric Topological Memory for Navigation (2018, methodology): “Gupta et al. (2017) build a 2D multi-scale metric map using the end-to-end trainable planning approach of Tamar et al. (2016).”
Zero-Shot Visual Imitation (2018, background): “Recent work in imitation learning (Duan et al., 2017; Finn et al., 2017; Gupta et al., 2017) can generalize to novel goals, but require a wealth of demonstrations comprised of expert state-actions for learning.”
Learning to Navigate Unseen Environments: Back Translation with Environmental Dropout (2019, methodology): “For example, Levine et al. (2016) used an end-to-end learned model to predict robotic actions from raw pixel data, Gupta et al. (2017) learned to navigate via mapping and planning, Sadeghi and Levine (2017) trained an agent to ﬂy in simulation and show its performance in the real world, and Gandhi…”
Gibson Env: Real-World Perception for Embodied Agents (2018, background): “are of limited use when it comes to active agent. Similarly, video datasets [57,70,101] are pre-recorded and thus passive. Virtual environments have been a remedy for this, classically [68] and today [106,36,31,83,47,41,11,9,72, 8,98]. Computer games, e.g. Minecraft [49], Doom [51] and GTA5 [69] hav”
Search on the Replay Buffer: Bridging Planning and Reinforcement Learning (2019, methodology): “Similar algorithms have been proposed for visual navigation (Anderson et al., 2018; Gupta et al., 2017; Mirowski et al., 2016; Zhu et al., 2017).”