Learning to Navigate in Cities Without a Map

From statwiki
Jump to navigation Jump to search

Paper: Learning to Navigate in Cities Without a Map[1]

A video of the paper is available here[2].

Introduction

Navigation is an attractive topic in many research disciplines and technology related domains such as neuroscience and robotics. The majority of algorithms are based on the following steps.

1. Building an explicit map

2. Planning and acting using that map.

In this article, based on this fact that human can learn to navigate through cities without using any special tool such as maps or GPS, authors propose new methods to show that a neural network agent can do the same thing by using visual observations. To do so, an interactive environment using Google StreetView Images and a dual pathway agent architecture are designed. As shown in figure 1, some parts of environment are built using Google StreetView images of New York City (Times Square, Central Park) and London (St. Paul’s Cathedral). The green cone represents the agent’s location and orientation. Although learning to navigate using visual aids is shown to be successful in some domains such as games and simulated environments using deep reinforcement learning (RL), it suffers from data inefficiency and sensitivity to changes in environment. Thus, it is unclear whether this method could be used for large-scale navigation. That’s why it became the subject of investigation in this paper.

Figure 1. Our environment is built of real-world places from StreetView. The figure shows diverse views and corresponding local maps in New York City (Times Square, Central Park) and London (St. Paul’s Cathedral). The green cone represents the agent’s location and orientation.

Related Works

Contribution

This paper has made the following contributions.

1. Designing a dual pathway agent. This agent can through a real city.

2. Using Goal-dependent learning. This means that the policy and value functions must adapt themselves to a sequence of goals that are provided as our input.

3. Leveraging a recurrent neural architecture. Using that, not only could navigation through a city be possible, but also the model is scalable for navigation in new cities.

4. Using a new environment which is built on top of Google StreetView. This provides real-world images for agent’s observation. Using this environment, agent should navigate from an arbitrary starting point to a goal and then to another goal etc. Also, London, Paris, and New York City are chosen for navigation.

Environment

Google StreetView consists of both high-resolution 360 degree imagery and graph connectivity. Also, it provides a public API. These features make it a valuable resource. In this work, large areas of New York, Paris, and London that contain between 7,000 and 65,500 nodes (and between 7,200 and 128,600 edges, respectively), have a mean node spacing of 10m, and cover a range of up to 5km chosen (Figure 2), without simplifying the underlying conncections. Also, the agent only sees RGB images that are visible in StreetView images (Figure 1).

Figure 2. Map of the 5 environments in New York City; our experiments focus on the NYU area as well as on transfer learning from the other areas to Wall Street (see Section 5.3). In the zoomed in area, each green dot corresponds to a unique panorama, the goal is marked in blue, and landmark locations are marked with red pins.

Agent Interface and the Courier Task

In RL environment, we need to define observations and actions in addition to taks. Xt and gt are inputs. Also, a first person view of 3D environment is simulated using cropped 60 degree square RGB image that is scaled to 84*84 pixels. Furthermore, the action space consists of 5 movements: “slow” rotate left or right (±22:5), “fast” rotate left or right (±67.5), or move forward.

There are lots of ways to specify the goal to the agent. In this paper, the current goal is chosen to be represented in terms of its proximity to a set L of fixed landmarks [math]\displaystyle{ L={(Lat_k, Long_k)} }[/math] which is specified using Latitude and Longitude coordinate system. For distance to the [math]\displaystyle{ k_{th} }[/math] landmark [math]\displaystyle{ {(d_{(t,k)}^g})_k }[/math] the goal vector contains [math]\displaystyle{ g_{(t,i)}=\tfrac{exp⁡(-αd_{(t,i)}^g)}{∑_k exp⁡(-αd_{(t,k)}^g)} }[/math]for [math]\displaystyle{ i_{th} }[/math] landmark with [math]\displaystyle{ α=0.002 }[/math] (Figure 3).

Figure 3. We illustrate the goal description by showing a goal and a set of 5 landmarks that are nearby, plus 4 that are more distant. The code [math]\displaystyle{ g_i }[/math] is a vector with a softmax-normalised distance to each landmark.

This form of representation has 2 advantages:

1. It could be extended to new environments easily.

2. It is intuitive. Even humans and animals use landmarks to be able to move from one place to another.

In this work 644 landmarks for NewYork, Paris, and London is manually defined. Furthermore, in each episode,which consists of 1000 steps, the agent starts from a random place with random orientation. when agent gets within 100 meter of goal, the next goal is randomly chosen. Finally the goal is proportional to the shortest path between agent and goal.

Methods

Goal-dependent Actor-Critic Reinforcement Learning

In this paper, the learning problem is based on Markov Decision Process, with state space S, action space A, environment Ɛ, and a set of possible goals G. The reward function depends on the current goal and state: [math]\displaystyle{ R: S×G×A → R }[/math]. maximize the expected sum of discounted rewards starting from state [math]\displaystyle{ s_0 }[/math] with discount Ƴ. Also the expected return from [math]\displaystyle{ s_t }[/math] depends on the goals that are sampled. So, policy and value functions are as follows.

\begin{align} g_t:π(α|s,g)=Pr(α_t=α|s_t=s, g_t=g) \end{align}

\begin{align} V^π(s,g)=E[R_t]=E[Σ_{k=0}^∞Ƴ^kr_{t+k}|s_t=s, g_t=g] \end{align}

Also an architecture with multiple pathways is designed to support two types of learning that is required for this problem. First, an agent needs an internal representation which is general and gives and understanding of a scene. Second, the agent needs to remember features that are available in a specific place.

Architectures

Figure 4. Comparison of architectures. Left: GoalNav is a convolutional encoder plus policy LSTM with goal description input. Middle: CityNav is a single-city navigation architecture with a separate goal LSTM and optional auxiliary heading (θ). Right: MultiCityNav is a multi-city architecture with individual goal LSTM pathways for each city.

The agent takes image pixels as input. Then, These pixels are passed through a convolutional network. The output of Convolutioin network is fed to a Long Short-Term Memory (LSTM) as well as the past reward [math]\displaystyle{ r_{t-1} }[/math] and previous action [math]\displaystyle{ α_{t-1} }[/math].

Three different architectures are described below.

The GoalNav architecture (Fig. 4a) which consists of a convolutional architecture and policy LSTM. Goal description [math]\displaystyle{ g_t }[/math], previous action, and reward are the inputs of this LSTM.

The CityNav architecture (Fig. 4b) consists of the previous architecture alongside an additional LSTM, called the goal LSTM. Inputs of this LSTM are visual features and the goal description. The CityNav agent also adds an auxiliary heading (\teta) prediction task which is defined as an angle between the north direction and the agent’s pose. This auxiliary task can speed up learning and provides relevant information.

The MultiCityNav architecture (Fig. 4c) is an extention of City-Nav for learning in different cities. This is done using parallel connection of goal LSTMs for encapsulating locale-specific features, for each city. Moreover,the convolutional architecture and the policy LSTM become general after training on a number of cities.So, new goal LSTMs are required to be trained for new cities.

Curriculum Learning