IEEE Robotics & Automation Magazine - June 2020 - 48
1) Observation stacking policies [7]: stacking the last N observations (o t, o t - 1, f o N ) and using this stack as the input of
the network
2) Recurrent policies [13]: including RNN layers in the policy
architecture.
One of the main issues with observation stacking is that
the memory of these models is determined by the number of
stacked observations. The overhead increases rapidly for larger sequences in high-dimensional observation problems.
In contrast, RNNs can model information for an arbitrarily
long period of time [14]. Also, they do not add input-related
overheads because, when these models are evaluated, they use
only the last observation. Therefore, RNNs have a lower computational cost than observation stacking. Given the more
practical usage of recurrent models and their capability of representing arbitrarily long sequences, in this article, we use
RNN-based policies (with LSTM layers) in the proposed NN
architecture. Nevertheless, the use of LSTMs has a critical disadvantage since their training is more complex and requires
more data, something very problematic when considering
human teachers and real systems. We now introduce SRL,
which helps to accelerate LSTM convergence.
SRL
In most of the problems faced in robotics, the state s t, which
fully describes the situation of the environment at time step t,
is not fully accessible from the robot's observation o t . As
mentioned, in several problems these observations lack the
temporal information required in the state description. Moreover, these observations tend to be raw sensor measurements
that can be high-dimensional, highly redundant, and ambiguous. A portion of this data may even be irrelevant.
As a consequence, to successfully solve these problems a
policy needs to 1) find temporal correlations between several
consecutive observations and 2) extract relevant features
from observations that are hard to interpret. However, finding relations among these large data structures with the
underlying phenomena of the environment while also learning controllers can be extremely inefficient. Therefore, efficiently building controllers on top of raw observations
requires learning-informative, low-dimensional SRs [15].
The objective of SRL is to obtain an observer capable of generating such representations.
Algorithm 1: (HG-)DAgger
1: Require: demonstrations database D with initial
demonstrations, policy update frequency b
2: for t = 1, 2, f do
3:
if mod(t, b) is 0 then
4:
update r i from D
5:
observe state o t
6:
select action from agent or expert
7:
execute action
8:
feedback provide label a )t for o t, if necessary
9:
aggregate (o t, a )t ) to D
48
*
IEEE ROBOTICS & AUTOMATION MAGAZINE
*
JUNE 2020
A compact representation of a state is considered suitable
for control if the resulting SR
● is Markovian
● has good generalization to unseen states
● is defined in low-dimensional space (considerably lower
than the actual observation dimensionality) [9].
Along with the control objective function (e.g., reward
function and imitation cost function), other objective functions can be used for SRL [10]:
● observation reconstruction
● forward model or next observation prediction
● the inverse model
● the reward function
● the value function.
Interactive Learning Methods
This section briefly introduces two approaches for interactively learning from human teachers while agents are executing a task.
Data Aggregation: DAgger and Human-Gated DAgger
DAgger [12] is an IIL algorithm that aims to collect data
through online sampling. To achieve this, trajectories are generated by combining the agent's policy r i and the expert's
policy. The observations o t and the demonstrator's corresponding actions a )t are paired and added to a database D,
which is used for training the policy's parameters i iteratively
in a supervised learning manner to asymptotically approach
the expert's policy. At the beginning of the learning process,
the demonstrator has all the influence over the trajectory
made by the agent; then the probability of following the demonstrator's actions decays exponentially.
For working in real-world systems with humans as demonstrators, a variation of DAgger, human-gated DAgger
(HG-DAgger) [2], was introduced. In this approach, the
demonstrator is not expected to give labels across every action
of the agent but only in places where she/he considers that the
agent's policy needs improvement. Only these labels are
aggregated to the database and used for updating the policy.
Additionally, every time feedback is given by the human, the
policy will follow the provided action. As a safety measure,
in HG-Dagger, the uncertainty of the policy across the
observation space is estimated; that element is omitted in
this article. Algorithm 1 shows the general structure of
DAgger and HG-DAgger.
D-COACH
In this framework, humans shape policies, giving occasional
corrective feedback for actions executed by the agents [11].
The human indicates agent actions that she/he considers to be
erroneous through a binary signal h t, the direction in which
the action should be modified. This feedback is used to generate an error signal for updating the policy parameters i. It is
performed in a supervised learning manner, with the cost
function J and using the mean squared error and stochastic
gradient descent. Hence, the update rule is
IEEE Robotics & Automation Magazine - June 2020
Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2020
Contents
IEEE Robotics & Automation Magazine - June 2020 - Cover1
IEEE Robotics & Automation Magazine - June 2020 - Cover2
IEEE Robotics & Automation Magazine - June 2020 - Contents
IEEE Robotics & Automation Magazine - June 2020 - 2
IEEE Robotics & Automation Magazine - June 2020 - 3
IEEE Robotics & Automation Magazine - June 2020 - 4
IEEE Robotics & Automation Magazine - June 2020 - 5
IEEE Robotics & Automation Magazine - June 2020 - 6
IEEE Robotics & Automation Magazine - June 2020 - 7
IEEE Robotics & Automation Magazine - June 2020 - 8
IEEE Robotics & Automation Magazine - June 2020 - 9
IEEE Robotics & Automation Magazine - June 2020 - 10
IEEE Robotics & Automation Magazine - June 2020 - 11
IEEE Robotics & Automation Magazine - June 2020 - 12
IEEE Robotics & Automation Magazine - June 2020 - 13
IEEE Robotics & Automation Magazine - June 2020 - 14
IEEE Robotics & Automation Magazine - June 2020 - 15
IEEE Robotics & Automation Magazine - June 2020 - 16
IEEE Robotics & Automation Magazine - June 2020 - 17
IEEE Robotics & Automation Magazine - June 2020 - 18
IEEE Robotics & Automation Magazine - June 2020 - 19
IEEE Robotics & Automation Magazine - June 2020 - 20
IEEE Robotics & Automation Magazine - June 2020 - 21
IEEE Robotics & Automation Magazine - June 2020 - 22
IEEE Robotics & Automation Magazine - June 2020 - 23
IEEE Robotics & Automation Magazine - June 2020 - 24
IEEE Robotics & Automation Magazine - June 2020 - 25
IEEE Robotics & Automation Magazine - June 2020 - 26
IEEE Robotics & Automation Magazine - June 2020 - 27
IEEE Robotics & Automation Magazine - June 2020 - 28
IEEE Robotics & Automation Magazine - June 2020 - 29
IEEE Robotics & Automation Magazine - June 2020 - 30
IEEE Robotics & Automation Magazine - June 2020 - 31
IEEE Robotics & Automation Magazine - June 2020 - 32
IEEE Robotics & Automation Magazine - June 2020 - 33
IEEE Robotics & Automation Magazine - June 2020 - 34
IEEE Robotics & Automation Magazine - June 2020 - 35
IEEE Robotics & Automation Magazine - June 2020 - 36
IEEE Robotics & Automation Magazine - June 2020 - 37
IEEE Robotics & Automation Magazine - June 2020 - 38
IEEE Robotics & Automation Magazine - June 2020 - 39
IEEE Robotics & Automation Magazine - June 2020 - 40
IEEE Robotics & Automation Magazine - June 2020 - 41
IEEE Robotics & Automation Magazine - June 2020 - 42
IEEE Robotics & Automation Magazine - June 2020 - 43
IEEE Robotics & Automation Magazine - June 2020 - 44
IEEE Robotics & Automation Magazine - June 2020 - 45
IEEE Robotics & Automation Magazine - June 2020 - 46
IEEE Robotics & Automation Magazine - June 2020 - 47
IEEE Robotics & Automation Magazine - June 2020 - 48
IEEE Robotics & Automation Magazine - June 2020 - 49
IEEE Robotics & Automation Magazine - June 2020 - 50
IEEE Robotics & Automation Magazine - June 2020 - 51
IEEE Robotics & Automation Magazine - June 2020 - 52
IEEE Robotics & Automation Magazine - June 2020 - 53
IEEE Robotics & Automation Magazine - June 2020 - 54
IEEE Robotics & Automation Magazine - June 2020 - 55
IEEE Robotics & Automation Magazine - June 2020 - 56
IEEE Robotics & Automation Magazine - June 2020 - 57
IEEE Robotics & Automation Magazine - June 2020 - 58
IEEE Robotics & Automation Magazine - June 2020 - 59
IEEE Robotics & Automation Magazine - June 2020 - 60
IEEE Robotics & Automation Magazine - June 2020 - 61
IEEE Robotics & Automation Magazine - June 2020 - 62
IEEE Robotics & Automation Magazine - June 2020 - 63
IEEE Robotics & Automation Magazine - June 2020 - 64
IEEE Robotics & Automation Magazine - June 2020 - 65
IEEE Robotics & Automation Magazine - June 2020 - 66
IEEE Robotics & Automation Magazine - June 2020 - 67
IEEE Robotics & Automation Magazine - June 2020 - 68
IEEE Robotics & Automation Magazine - June 2020 - 69
IEEE Robotics & Automation Magazine - June 2020 - 70
IEEE Robotics & Automation Magazine - June 2020 - 71
IEEE Robotics & Automation Magazine - June 2020 - 72
IEEE Robotics & Automation Magazine - June 2020 - 73
IEEE Robotics & Automation Magazine - June 2020 - 74
IEEE Robotics & Automation Magazine - June 2020 - 75
IEEE Robotics & Automation Magazine - June 2020 - 76
IEEE Robotics & Automation Magazine - June 2020 - 77
IEEE Robotics & Automation Magazine - June 2020 - 78
IEEE Robotics & Automation Magazine - June 2020 - 79
IEEE Robotics & Automation Magazine - June 2020 - 80
IEEE Robotics & Automation Magazine - June 2020 - 81
IEEE Robotics & Automation Magazine - June 2020 - 82
IEEE Robotics & Automation Magazine - June 2020 - 83
IEEE Robotics & Automation Magazine - June 2020 - 84
IEEE Robotics & Automation Magazine - June 2020 - 85
IEEE Robotics & Automation Magazine - June 2020 - 86
IEEE Robotics & Automation Magazine - June 2020 - 87
IEEE Robotics & Automation Magazine - June 2020 - 88
IEEE Robotics & Automation Magazine - June 2020 - 89
IEEE Robotics & Automation Magazine - June 2020 - 90
IEEE Robotics & Automation Magazine - June 2020 - 91
IEEE Robotics & Automation Magazine - June 2020 - 92
IEEE Robotics & Automation Magazine - June 2020 - 93
IEEE Robotics & Automation Magazine - June 2020 - 94
IEEE Robotics & Automation Magazine - June 2020 - 95
IEEE Robotics & Automation Magazine - June 2020 - 96
IEEE Robotics & Automation Magazine - June 2020 - 97
IEEE Robotics & Automation Magazine - June 2020 - 98
IEEE Robotics & Automation Magazine - June 2020 - 99
IEEE Robotics & Automation Magazine - June 2020 - 100
IEEE Robotics & Automation Magazine - June 2020 - 101
IEEE Robotics & Automation Magazine - June 2020 - 102
IEEE Robotics & Automation Magazine - June 2020 - 103
IEEE Robotics & Automation Magazine - June 2020 - 104
IEEE Robotics & Automation Magazine - June 2020 - 105
IEEE Robotics & Automation Magazine - June 2020 - 106
IEEE Robotics & Automation Magazine - June 2020 - 107
IEEE Robotics & Automation Magazine - June 2020 - 108
IEEE Robotics & Automation Magazine - June 2020 - 109
IEEE Robotics & Automation Magazine - June 2020 - 110
IEEE Robotics & Automation Magazine - June 2020 - 111
IEEE Robotics & Automation Magazine - June 2020 - 112
IEEE Robotics & Automation Magazine - June 2020 - 113
IEEE Robotics & Automation Magazine - June 2020 - 114
IEEE Robotics & Automation Magazine - June 2020 - 115
IEEE Robotics & Automation Magazine - June 2020 - 116
IEEE Robotics & Automation Magazine - June 2020 - 117
IEEE Robotics & Automation Magazine - June 2020 - 118
IEEE Robotics & Automation Magazine - June 2020 - 119
IEEE Robotics & Automation Magazine - June 2020 - 120
IEEE Robotics & Automation Magazine - June 2020 - 121
IEEE Robotics & Automation Magazine - June 2020 - 122
IEEE Robotics & Automation Magazine - June 2020 - 123
IEEE Robotics & Automation Magazine - June 2020 - 124
IEEE Robotics & Automation Magazine - June 2020 - 125
IEEE Robotics & Automation Magazine - June 2020 - 126
IEEE Robotics & Automation Magazine - June 2020 - 127
IEEE Robotics & Automation Magazine - June 2020 - 128
IEEE Robotics & Automation Magazine - June 2020 - 129
IEEE Robotics & Automation Magazine - June 2020 - 130
IEEE Robotics & Automation Magazine - June 2020 - 131
IEEE Robotics & Automation Magazine - June 2020 - 132
IEEE Robotics & Automation Magazine - June 2020 - 133
IEEE Robotics & Automation Magazine - June 2020 - 134
IEEE Robotics & Automation Magazine - June 2020 - 135
IEEE Robotics & Automation Magazine - June 2020 - 136
IEEE Robotics & Automation Magazine - June 2020 - 137
IEEE Robotics & Automation Magazine - June 2020 - 138
IEEE Robotics & Automation Magazine - June 2020 - 139
IEEE Robotics & Automation Magazine - June 2020 - 140
IEEE Robotics & Automation Magazine - June 2020 - 141
IEEE Robotics & Automation Magazine - June 2020 - 142
IEEE Robotics & Automation Magazine - June 2020 - 143
IEEE Robotics & Automation Magazine - June 2020 - 144
IEEE Robotics & Automation Magazine - June 2020 - 145
IEEE Robotics & Automation Magazine - June 2020 - 146
IEEE Robotics & Automation Magazine - June 2020 - 147
IEEE Robotics & Automation Magazine - June 2020 - 148
IEEE Robotics & Automation Magazine - June 2020 - 149
IEEE Robotics & Automation Magazine - June 2020 - 150
IEEE Robotics & Automation Magazine - June 2020 - 151
IEEE Robotics & Automation Magazine - June 2020 - 152
IEEE Robotics & Automation Magazine - June 2020 - 153
IEEE Robotics & Automation Magazine - June 2020 - 154
IEEE Robotics & Automation Magazine - June 2020 - 155
IEEE Robotics & Automation Magazine - June 2020 - 156
IEEE Robotics & Automation Magazine - June 2020 - 157
IEEE Robotics & Automation Magazine - June 2020 - 158
IEEE Robotics & Automation Magazine - June 2020 - 159
IEEE Robotics & Automation Magazine - June 2020 - 160
IEEE Robotics & Automation Magazine - June 2020 - 161
IEEE Robotics & Automation Magazine - June 2020 - 162
IEEE Robotics & Automation Magazine - June 2020 - 163
IEEE Robotics & Automation Magazine - June 2020 - 164
IEEE Robotics & Automation Magazine - June 2020 - 165
IEEE Robotics & Automation Magazine - June 2020 - 166
IEEE Robotics & Automation Magazine - June 2020 - 167
IEEE Robotics & Automation Magazine - June 2020 - 168
IEEE Robotics & Automation Magazine - June 2020 - 169
IEEE Robotics & Automation Magazine - June 2020 - 170
IEEE Robotics & Automation Magazine - June 2020 - 171
IEEE Robotics & Automation Magazine - June 2020 - 172
IEEE Robotics & Automation Magazine - June 2020 - 173
IEEE Robotics & Automation Magazine - June 2020 - 174
IEEE Robotics & Automation Magazine - June 2020 - 175
IEEE Robotics & Automation Magazine - June 2020 - 176
IEEE Robotics & Automation Magazine - June 2020 - 177
IEEE Robotics & Automation Magazine - June 2020 - 178
IEEE Robotics & Automation Magazine - June 2020 - 179
IEEE Robotics & Automation Magazine - June 2020 - 180
IEEE Robotics & Automation Magazine - June 2020 - 181
IEEE Robotics & Automation Magazine - June 2020 - 182
IEEE Robotics & Automation Magazine - June 2020 - 183
IEEE Robotics & Automation Magazine - June 2020 - 184
IEEE Robotics & Automation Magazine - June 2020 - Cover3
IEEE Robotics & Automation Magazine - June 2020 - Cover4
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_december2023
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_september2023
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_june2023
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_march2023
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_december2022
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_september2022
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_june2022
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_march2022
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_december2021
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_september2021
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_june2021
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_march2021
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_december2020
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_september2020
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_june2020
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_march2020
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_december2019
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_september2019
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_june2019
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_march2019
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_december2018
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_september2018
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_june2018
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_march2018
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_december2017
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_september2017
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_june2017
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_march2017
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_december2016
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_september2016
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_june2016
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_march2016
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_december2015
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_september2015
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_june2015
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_march2015
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_december2014
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_september2014
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_june2014
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_march2014
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_december2013
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_september2013
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_june2013
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_march2013
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_december2012
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_september2012
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_june2012
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_march2012
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_december2011
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_september2011
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_june2011
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_march2011
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_december2010
https://www.nxtbook.com/nxtbooks/ieee/roboticsautomation_september2010
https://www.nxtbookmedia.com