-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathreport.lyx
219 lines (176 loc) · 4.55 KB
/
report.lyx
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
#LyX 2.1 created this file. For more info see http://www.lyx.org/
\lyxformat 474
\begin_document
\begin_header
\textclass article
\use_default_options true
\maintain_unincluded_children false
\language english
\language_package default
\inputencoding auto
\fontencoding global
\font_roman default
\font_sans default
\font_typewriter default
\font_math auto
\font_default_family default
\use_non_tex_fonts false
\font_sc false
\font_osf false
\font_sf_scale 100
\font_tt_scale 100
\graphics default
\default_output_format default
\output_sync 0
\bibtex_command default
\index_command default
\paperfontsize default
\use_hyperref false
\papersize default
\use_geometry false
\use_package amsmath 1
\use_package amssymb 1
\use_package cancel 1
\use_package esint 1
\use_package mathdots 1
\use_package mathtools 1
\use_package mhchem 1
\use_package stackrel 1
\use_package stmaryrd 1
\use_package undertilde 1
\cite_engine basic
\cite_engine_type default
\biblio_style plain
\use_bibtopic false
\use_indices false
\paperorientation portrait
\suppress_date false
\justification true
\use_refstyle 1
\index Index
\shortcut idx
\color #008000
\end_index
\secnumdepth 3
\tocdepth 3
\paragraph_separation indent
\paragraph_indentation default
\quotes_language english
\papercolumns 1
\papersides 1
\paperpagestyle default
\tracking_changes false
\output_changes false
\html_math_output 0
\html_css_as_file 0
\html_be_strict false
\end_header
\begin_body
\begin_layout Title
Reinforcement learning to solve
\begin_inset CommandInset href
LatexCommand href
name "Agar.io"
target "agar.io"
\end_inset
\end_layout
\begin_layout Author
\begin_inset CommandInset href
LatexCommand href
name "Vishruit"
target "vishruit@gmail.com"
type "mailto:"
\end_inset
,
\begin_inset CommandInset href
LatexCommand href
name "Suhas"
target "suhas.gundimeda@gmail.com"
type "mailto:"
\end_inset
\end_layout
\begin_layout Abstract
Agar.io is a blob-eat-blob 2D world where the player controls a circular
blob whose primary objective is to accumulate the largest amount of volume.
This objective boils down to the amount of food it retains, supply of which
consists of both static randomly dropped food and other real-time players.
The available high-level decisions and actions for the blob include eating
static food, actively avoiding bigger blobs, actively pursuing them.
In addition, the blob has the additional possible action of splitting its
mass to project a percentage of itself with a higher velocity in the intended
direction.
\begin_inset Newline newline
\end_inset
The current report focuses on an exercise to solve a simpler version of
the problem
\begin_inset CommandInset citation
LatexCommand cite
key "key-1"
\end_inset
, a single player world with limited observability and continuous state
space.
\begin_inset Foot
status open
\begin_layout Plain Layout
Project repository, including latest report and source code at
\begin_inset CommandInset href
LatexCommand href
target "https://github.com/snugghash/verbose-spork"
\end_inset
\end_layout
\end_inset
\end_layout
\begin_layout Section
Problem statement
\end_layout
\begin_layout Standard
The problem space consists of a large grid-world where the agent has a limited
field of view, and their only primitive actions are forward movements or
rotations.
The objective is to keep moving to grids which have an green(positive reward)
object, and avoid red(negative reward) objects.
New objects are placed at random locations, keeping the total number of
each kind same.
The game isn't episodic, and the objective is to maximize reward/step averaged
over a thousand steps.
\end_layout
\begin_layout Standard
The agent has sensors which output the (continuous dimension) distance to
green, red and wall objects, along the single direction they sense.
\end_layout
\begin_layout Section
Formulation
\end_layout
\begin_layout Standard
We used parametrized state spaces and function approximation to store and
represent, as action values, the policy to follow if greedy.
\end_layout
\begin_layout Standard
Exploration was epsilon greedy.
\end_layout
\begin_layout Section
Results
\end_layout
\begin_layout Standard
\begin_inset Float figure
wide false
sideways false
status open
\begin_layout Plain Layout
\begin_inset Caption Standard
\begin_layout Plain Layout
Average reward over 100000 steps
\end_layout
\end_inset
\end_layout
\end_inset
\end_layout
\begin_layout Bibliography
\begin_inset CommandInset bibitem
LatexCommand bibitem
key "key-1"
\end_inset
http://cs.stanford.edu/people/karpathy/convnetjs/demo/rldemo.html
\end_layout
\end_body
\end_document