Batch Criteria

Block Density
Swarm Population Dynamics
Block Quantity
Block Motion Dynamics
Oracle
Task Allocation Policy
Temporal Variance

Swarm Population Dynamics

Cmdline Syntax

population_dynamics.C{cardinality}.F{Factor}[.{dynamics_type}{prob}[...]]

cardinality - The # of different values of each of the specified dynamics types to to test with (starting with the one on the cmdline). This defines the cardinality of the batched experiment.
Factor - The factor by which the starting value of all dynamics types specified on the cmdline are increased for each each experiment (i.e., value in last experiment in batch will be <start value> + cardinality; a linear increase).
dynamics_type - [B,D,M,R]
- B - Adds birth dynamics to the population. Has no effect by itself, as it specifies a pure birth process with \(\lambda=\infty\), \(\mu_{b}\) (a queue with an infinite # of robots in it which robots periodically leave from), resulting in dynamic swarm sizes which will increase from N…N over time. Can be specified with along with D|M|R, in which case swarm sizes will increase according to the birth rate up until N, given N robots at the start of simulation.
- D - Adds death dynamics to the population. By itself, it specifies a pure death process with \(\lambda_{d}=prob\), and \(\mu_{d}=\infty\) (a queue which robots enter but never leave), resulting in dynamic swarm sizes which will decrease from N…1 over time. Can be specified along with B|M|R.
- M|R - Adds malfunction/repair dynamics to the population. If M dynamics specified, R dynamics must also be specified, and vice versa. Together they specify the dynamics of the swarm as robots temporarily fail, and removed from simulation, and then later are re-introduced after they have been repaired (a queue with \(\lambda_{m}\) arrival rate and \(\mu_{r}\) repair rate). Can be specified along with B|D.

Important

The specified \(\lambda\) or \(\mu\) are the rate parameters of the exponential distribution used to distribute the event times of the Poisson process governing swarm sizes, NOT Poisson process parameter, which is their mean; e.g., \(\lambda=\frac{1}{\lambda_{d}}\) for death dynamics.

Examples:

population_dynamics.C10.F2p0.B0p001: 10 levels of population variability applied using a pure birth process with a 0.001 parameter, which will be linearly varied in [0.001,0.001*2.0*10]. For all experiments, the initial swarm is not controlled directly; the value in template input file will be used if swarm size is not set by another variable.
population_dynamics.C4.F3p0.D0p001: 4 levels of population variability applied using a pure death process with a 0.001 parameter, which will be linearly varied in [0.001,0.001*3.0*4]. For all experiments, the initial swarm size is not controlled directly; the value in template input file will be used if swarm size is not set by another variable.
population_dynamics.C8.F4p0.B0p001.D0p005: 8 levels of population variability applied using a birth-death process with a 0.001 parameter for birth and a 0.005 parameter for death, which will be linearly varied in [0.001,0.001*4.0*8] and [0.005, 0.005*4.0*8] respectively. For all experiments, the initial swarm is not controlled directly; the value in the template input file will be used if swarm size is is not set by another variable.
population_dynamics.C2.F1p5.M0p001.R0p005: 2 levels of population variability applied using a malfunction-repair process with a 0.001 parameter for malfunction and a 0.005 parameter for repair which will be linearly varied in [0.001, 0.001*1.5*2] and [0.005, 0.005*1.5*2] respectively. For all experiments, the initial swarm size is not controlled directly; the value in the template input file will be used if swarm size is not set by another variable.

Block Quantity

Cmdline Syntax

block_quantity.{block_type}.{increment_type}{N}

block_type - cube or ramp, depending on what type of blocks you want to control the count of.
increment_type - {Log,Linear}. If Log, then swarm sizes for each experiment are distributed 1…N by powers of 2. If Linear then block counts for each experiment are distributed linearly between 1…N, split evenly into 10 different sizes.
N - The maximum block count.

Examples:

block_quantity.cube.Log1024: Cube block counts 1…1024
block_quantity.ramp.Linear1000: Ramp block counts 100…1000

Block Density

Cmdline Syntax

block_density.CD{density}.I{Arena Size Increment}.C{cardinality}

density - <integer>p<integer> (i.e. 5p0 for 5.0)
Arena Size Increment - Size in meters that the X and Y dimensions should
increase by in between experiments. Larger values here will result in larger arenas and more blocks. Must be an integer.
cardinality How many experiments should be generated?

Examples:

block_density.CD1p0.I16.C4: Constant density of 1.0. Arena dimensions will increase by 16 in both X and Y for each experiment in the batch (4 total).

Block Motion Dynamics

Cmdline Syntax

block_motion_dynamics.C{cardinality}.F{Factor}.{dynamics_type}{prob}

cardinality - The # of different values of each of the specified dynamics types to to test with (starting with the one on the cmdline). This defines the cardinality of the batched experiment.
Factor - The factor by which the starting value of all dynamics types specified on the cmdline are increased for each each experiment (i.e., value in last experiment in batch will be <start value> + cardinality; a linear increase).
dynamics_type - [RW]
- RW - Adds random walk dynamics to the arena. Free blocks will execute a random walk with a specified probability each timestep.

Examples:

block_motion_dynamics.C10.F2p0.RW0p001: 10 levels of block motion variability applied using a random walk with a 0.001 probability for each block each timestep, which will be linearly varied in [0.001,0.001*2.0*10]. For all experiments, the initial swarm is not controlled directly; the value in template input file will be used if swarm size is not set by another variable.

Oracle

Cmdline Syntax

oracle.{oracle_name}[.Z{population}]

oracle_name - {entities, tasks}
- entities - Inject perfect information about locations about entities in the arena, such as blocks and caches.
- tasks - Inject perfect information about task execution and interface times.
population - Static size of the swarm to use (optional).

Examples:

oracle.entities.Z16 - All permutations of oracular information about entities in the arena, run with swarms of size 16.
oracle.tasks.Z8 - All permutations of oracular information about tasks in the arena, run with swarms of size 8.
oracle.entities - All permuntations of oracular information of entities in the arena (swarm size is not modified).

Task Allocation Policy

Cmdline Syntax

ta_policy_set.all[.Z{population}]

population - The swarm size to use (optional)

Examples:

ta_policy_set.all.Z16: All possible task allocation policies with swarms of size 16.
ta_policy_set.all: All possible task allocation policies; swarm size not modified.

Temporal Variance

Injecting waveforms into the swarm’s environment which affect the individual robot behavior to simulate changing outdoor conditions, changing object sizes/weights, etc.

Note

The graphs generated from this criteria exclude exp0.

Warning

Some of the temporal variance config is very FORDYCA specific; hopefully this will change in the future, or be pushed down to a project-specific extension of a base flexibility class.

Cmdline Syntax

temporal_variance.{variance_type}{waveform_type}[step_time][.Z{population}]

variance_type - [BC,BM,M].
- BC - Apply motion throttling to robot speed when it is carrying a block according to the specified waveform.
- BM - Apply the specified waveform when calculating robot block manipulation penalties (pickup, drop, etc.).
- M - Apply the specified waveform to robot motion unconditionally.
waveform_type - {Sine,Square,Sawtooth,Step{U,D},Constant}.
step_time - Timestep the step function should switch (optional).
population - The static swarm size to use (optional).

Examples:

temporal_variance.BCSine.Z16 - Block carry sinusoidal variance in a swarm of size 16.
temporal_variance.BCStep50000.Z32 - Block carry step variance switch at 50000 timesteps in a swarm of size 32.
temporal_variance.BCStep50000 - Block carry step variance switching at 50000 timesteps; swarm size not modified.

The frequency, amplitude, offset, and phase of the waveforms is set via the main.yaml configuration file for a project (not an easy way to specify ranges in a single batch criteria definition string). The relevant section is shown below.

For the {Sine,Square,Sawtooth} waveforms, the cardinality of the batched experiment is determined by: (Size of Hz list -1) * (Size of BC_amp/BM_amp list - 1).

YAML Config

perf:
  ...
  flexibility:
    # The range of Hz to use for generated waveforms. Applies to Sine,
    # Sawtooth, Square waves. There is no limit for the length of the list.
    hz:
      - frequency1
      - frequency2
      - frequency3
      - ...
    # The range of block manipulation penalties to use if that is the type of
    # applied temporal variance (BM). Specified in timesteps. There is no
    # limit for the length of the list.
    BM_amp:
      - penalty1
      - penalty2
      - penalty3
      - ...
   # The range of block carry penalties to use if that is the type of applied
   # temporal variance (BC). Specified as percent slowdown: [0.0, 1.0]. There
   #is no limit for the length of the list.
   BC_amp:
      - percent1
      - percent2
      - percent3
      - ...

   # The range of motion throttle penalties to use if that is the type of
   # applied temporal variance (M). Specified as percent slowdown: [0.0,
   # 1.0]. There is no limit for the length of the list.
   M_amp:
      - percent1
      - percent2
      - percent3
      - ...

Experiment Definitions

exp0 - Ideal conditions, which is a Constant waveform with amplitude BC_amp[0], BM_amp[0], M_amp[0] depending.
exp1-expN
- Cardinality of |hz| * |BM_amp| if the variance type is BM and the waveform type is Sine, Square, or Sawtooth.
- Cardinality of |hz| * |BC_amp| if the variance type is BC and the waveform type is Sine, Square, or Sawtooth.
- Cardinality of |hz| * |M_amp| if the variance type is M and the waveform type is Sine, Square, or Sawtooth.
- Cardinality of |BM_amp| if the variance type is BM and the waveform type is StepU, StepD.
- Cardinality of |BC_amp| if the variance type is BC and the waveform type is StepU, StepD.
- Cardinality of |M_amp| if the variance type is M and the waveform type is StepU, StepD.