You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Neurobiological theories of spatial cognition developed with respect to
17
-
recording data from relatively small and/or simplistic environments compared
18
-
to animals’ natural habitats. It has been unclear how to extend theoretical
19
-
models to large or complex spaces. Complementarily, in autonomous systems
20
-
technology, applications have been growing for distributed control methods
21
-
that scale to large numbers of low-footprint mobile platforms. Animals and
22
-
many-robot groups must solve common problems of navigating complex and
23
-
uncertain environments. Here, we introduce the NeuroSwarms control framework
24
-
to investigate whether adaptive, autonomous swarm control of minimal
25
-
artificial agents can be achieved by direct analogy to neural circuits of
26
-
rodent spatial cognition. NeuroSwarms analogizes agents to neurons and
27
-
swarming groups to recurrent networks. We implemented neuron-like agent
28
-
interactions in which mutually visible agents operate as if they were
29
-
reciprocally connected place cells in an attractor network. We attributed a
30
-
phase state to agents to enable patterns of oscillatory synchronization
31
-
similar to hippocampal models of theta-rhythmic (5–12 Hz) sequence
32
-
generation. We demonstrate that multi-agent swarming and reward-approach
33
-
dynamics can be expressed as a mobile form of Hebbian learning and that
34
-
NeuroSwarms supports a single-entity paradigm that directly informs
35
-
theoretical models of animal cognition. We present emergent behaviors
36
-
including phase-organized rings and trajectory sequences that interact with
37
-
environmental cues and geometry in large, fragmented mazes. Thus, NeuroSwarms
38
-
is a model artificial spatial system that integrates autonomous control and
39
-
theoretical neuroscience to potentially uncover common principles to advance
40
-
both domains.
16
+
> Neurobiological theories of spatial cognition developed with respect to
17
+
> recording data from relatively small and/or simplistic environments compared
18
+
> to animals’ natural habitats. It has been unclear how to extend theoretical
19
+
> models to large or complex spaces. Complementarily, in autonomous systems
20
+
> technology, applications have been growing for distributed control methods
21
+
> that scale to large numbers of low-footprint mobile platforms. Animals and
22
+
> many-robot groups must solve common problems of navigating complex and
23
+
> uncertain environments. Here, we introduce the NeuroSwarms control framework
24
+
> to investigate whether adaptive, autonomous swarm control of minimal
25
+
> artificial agents can be achieved by direct analogy to neural circuits of
26
+
> rodent spatial cognition. NeuroSwarms analogizes agents to neurons and
27
+
> swarming groups to recurrent networks. We implemented neuron-like agent
28
+
> interactions in which mutually visible agents operate as if they were
29
+
> reciprocally connected place cells in an attractor network. We attributed a
30
+
> phase state to agents to enable patterns of oscillatory synchronization
31
+
> similar to hippocampal models of theta-rhythmic (5–12 Hz) sequence
32
+
> generation. We demonstrate that multi-agent swarming and reward-approach
33
+
> dynamics can be expressed as a mobile form of Hebbian learning and that
34
+
> NeuroSwarms supports a single-entity paradigm that directly informs
35
+
> theoretical models of animal cognition. We present emergent behaviors
36
+
> including phase-organized rings and trajectory sequences that interact with
37
+
> environmental cues and geometry in large, fragmented mazes. Thus, NeuroSwarms
38
+
> is a model artificial spatial system that integrates autonomous control and
39
+
> theoretical neuroscience to potentially uncover common principles to advance
40
+
> both domains.
41
41
42
42
**Installation**
43
43
44
-
First, either set up a new python environment (with either venv or anaconda) and install the required dependencies. Using Anaconda, this would look like,
44
+
First, either set up a new python environment (with either venv or Anaconda) and
45
+
install the required dependencies. Using Anaconda, this would look like,
Then, in the top-level `neuroswarms` folder (either cloned or downloaded from this repository), you can do a developer install of neuroswarms if you are interested in working with the code. In your shell,
52
+
Then, in the top-level `neuroswarms` folder (either cloned or downloaded from
53
+
this repository), you can do a developer install (with the `-e` option below) if
54
+
you are interested in working with the code:
52
55
53
56
```bash
54
-
$ cd /path/to/neuroswarms
55
-
$ pip install -e .
56
-
```
57
+
(neuroswarms)$ cd /path/to/neuroswarms
58
+
(neuroswarms)$ pip install -e .
59
+
```
60
+
61
+
If you have the `mpv` video player installed, it will be used to automatically
62
+
play the movie file of the simulation once it is saved. You can install `mpv`
63
+
with `brew install mpv` (macOS) or `sudo apt install mpv` (linux).
64
+
65
+
**Usage**
66
+
67
+
An example script in `scripts/run-example.py` shows how to create a
68
+
`NeuroswarmsModel` object and call its `.simulate(...)` method with parameter
69
+
values to run a trial simulation.
70
+
71
+
The `mapdata` folder contains the precomputed data for the two environments
72
+
presented in the paper: the 'multi-reward arena' and the 'large hairpin maze'.
73
+
You can choose either of these environments by setting the `env` parameter to
0 commit comments