... | ... | @@ -210,21 +210,21 @@ A `CostFunction` element has the following **main attributes**: |
|
|
|
|
|
The following cost functions are currently supported:
|
|
|
|
|
|
| Name in XML | Description | Used in example policy | Notes |
|
|
|
| ------ | ------ | ------ | ------ |
|
|
|
| `FOEAvoidance` | An avoidance function based on the focus of expansion. | - | 1,2 |
|
|
|
| `GoalReachingForce` | The goal-reaching force described by Helbing and Molnár (1995). | SocialForces | 1 |
|
|
|
| `Karamouzas` | The cost function used in the velocity-based algorithm by Karamouzas and Overmars (2010). | Karamouzas | - |
|
|
|
| `Moussaid` | The cost function used in the velocity-based algorithm by Moussaid et al. (2011). | Moussaid | - |
|
|
|
| `ORCA` | The cost function used in the ORCA method by van den Berg et al. (2011). | ORCA | 2,4 |
|
|
|
| `Paris` | A cost function inspired by the collision-avoidance algorithm of Paris et al. (2007). | - | 2,3 |
|
|
|
| `PLEdestrians` | The cost function used in the PLEdestrians model by Guy et al. (2010). | PLEdestrians | 5 |
|
|
|
| `PowerLaw` | The force used in the 'Universal Power Law' model by Karamouzas et al. (2014) | PowerLaw | 1,2 |
|
|
|
| `RandomFunction` | A function that returns random costs and gradients. Useful for adding noise to the system. | - | - |
|
|
|
| `RVO` | The cost function used in the original RVO implementation by van den Berg et al. (2008). | RVO | - |
|
|
|
| `SocialForcesAvoidance` | The avoidance component of the social-force model by Helbing and Molnár (1995). | SocialForces | 1 |
|
|
|
| `TtcaDca` | The cost function used in the vision-based steering algorithm by Dutra et al. (2016), simplified to our domain. | Dutra | 1,2 |
|
|
|
| `VanToll` | A custom collision-avoidance cost function that is sometimes used by our own team. | VanToll | - |
|
|
|
| Name in XML | Description | Used in example policy | Additional attributes | Notes |
|
|
|
| ------ | ------ | ------ | ------ | ------ |
|
|
|
| `FOEAvoidance` | An avoidance function based on the focus of expansion. | - | - | 1,2 |
|
|
|
| `GoalReachingForce` | The goal-reaching force described by Helbing and Molnár (1995). | SocialForces | - | 1 |
|
|
|
| `Karamouzas` | The cost function used in the velocity-based algorithm by Karamouzas and Overmars (2010). | Karamouzas | `alpha`, `beta`, `gamma`, `delta`, `t_max` | - |
|
|
|
| `Moussaid` | The cost function used in the velocity-based algorithm by Moussaid et al. (2011). | Moussaid | `d_max` | - |
|
|
|
| `ORCA` | The cost function used in the ORCA method by van den Berg et al. (2011). | ORCA | `timeHorizon` | 2,4 |
|
|
|
| `Paris` | A cost function inspired by the collision-avoidance algorithm of Paris et al. (2007). | - | `w_a`, `w_b`, `t_max` | 2,3 |
|
|
|
| `PLEdestrians` | The cost function used in the PLEdestrians model by Guy et al. (2010). | PLEdestrians | `w_a`, `w_b`, `t_min`, `t_max` | 5 |
|
|
|
| `PowerLaw` | The force used in the 'Universal Power Law' model by Karamouzas et al. (2014) | PowerLaw | `k`, `tau0` | 1,2 |
|
|
|
| `RandomFunction` | A function that returns random costs and gradients. Useful for adding noise to the system. | - | - | - |
|
|
|
| `RVO` | The cost function used in the original RVO implementation by van den Berg et al. (2008). | RVO | `w` | - |
|
|
|
| `SocialForcesAvoidance` | The avoidance component of the social-force model by Helbing and Molnár (1995). | SocialForces | `dt`, `V0`, `sigma`, `U0`, `R` | 1 |
|
|
|
| `TtcaDca` | The cost function used in the vision-based steering algorithm by Dutra et al. (2016), simplified to our domain. | Dutra | `sigmaTtca`, `sigmaDca`, `sigmaAngle_goal`, `sigmaSpeed_goal` | 1,2 |
|
|
|
| `VanToll` | A custom collision-avoidance cost function that is sometimes used by our own team. | VanToll | `max_distance` | - |
|
|
|
|
|
|
*Notes:*
|
|
|
1. This cost function is meant to be used in a gradient-based policy. You are free to use it in a sampling-based policy as well, but it may not behave exactly as intended.
|
... | ... | @@ -233,4 +233,4 @@ The following cost functions are currently supported: |
|
|
4. This cost function also supports global optimization, using a direct port of the ORCA source code.
|
|
|
5. This cost function does not yet support global optimization, even though it is theoretically possible.
|
|
|
|
|
|
Each specific cost function can have its own **specific attributes**. We are currently not listing these attributes here. Instead, they are specified in the (Doxygen) comments of each cost function's C++ header file. For example, the documentation in `core/CostFunctions/RVO.h` describes the parameters of the RVO cost function. Therefore, the best option is to look in the Doxygen comments of the cost function that you want to use. |
|
|
\ No newline at end of file |
|
|
As seen in the table above, many cost functions have their own **additional attributes**. These are usually one-on-one translations of the algorithm's parameters described in the corresponding paper. If you want to know more about them, we recommend you to take a look at the (Doxygen) documentation of the UMANS codebase. For example, the documentation in `core/CostFunctions/SocialForcesAvoidance.h` describes the parameters of the `SocialForcesAvoidance` cost function in more detail. |
|
|
\ No newline at end of file |