Intelligent systems today, including physical robots, are capable of amazing feats compared to their early predecessors. However, much of the planning and control relies on the accurate identification of the robotic system, often in the form of known physics equations derived by experienced engineers. In the case when the system model is unknown, various function approximations have been used in the context of model-based reinforcement learning. There is a typical trade-off between the explicability and model capacity when applying function approximation, e.g., deep neural networks may model complex system dynamics but are hard to interpret and optimize. Typically, control and planning tasks require much higher accuracy compared to regular machine learning tasks. In addition, it naturally occurs that unseen parts of the state space are visited. So we wish to address the problem of learning models that are highly accurate and also able to extrapolate.
Instead of resorting to black-box function approximation we designed a system to identify the system equation from a limited data range using a gradient-based neural network training[ ]. In this way, we solve a symbolic regression problem using a continuous optimization method. The crucial aspect of the method is to find the simplest solution that still explains the data well - implementing Occam's razor.
The identified equations are suitable for extrapolation to unseen domains which we demonstrate using several synthetic and real-world systems. Most importantly we show how to use these learned equations to obtain a highly efficient learning algorithm for control and planning. In our first experiment, a cart-pendulum system learns to swing up from scratch after two random rollouts. Ongoing work is to improve the efficiency and applicability of the method and applying it to challenging robotic systems.
Our website uses cookies. Some of them are mandatory, while others allow us to improve your user experience on our website. The settings you have made can be edited at any time.
or
Essential
in2cookiemodal-selection
Required to save the user selection of the cookie settings.
3 months
be_lastLoginProvider
Required for the TYPO3 backend login to determine the time of the last login.
3 months
be_typo_user
This cookie tells the website whether a visitor is logged into the TYPO3 backend and has the rights to manage it.
Browser session
ROUTEID
These cookies are set to always direct the user to the same server.
Browser session
fe_typo_user
Enables frontend login.
Browser session
Videos
iframeswitch
Used to show all third-party contents.
3 months
yt-player-bandaid-host
Is used to display YouTube videos.
Persistent
yt-player-bandwidth
Is used to determine the optimal video quality based on the visitor's device and network settings.
Persistent
yt-remote-connected-devices
Saves the settings of the user's video player using embedded YouTube video.
Persistent
yt-remote-device-id
Saves the settings of the user's video player using embedded YouTube video.
Persistent
yt-player-headers-readable
Collects data about visitors' interaction with the site's video content - This data is used to make the site's video content more relevant to the visitor.
Persistent
yt-player-volume
Is used to save volume preferences for YouTube videos.
Persistent
yt-player-quality
Is used to save the quality settings for YouTube videos.
Persistent
yt-remote-session-name
Saves the settings of the user's video player using embedded YouTube video.
Browser session
yt-remote-session-app
Saves the settings of the user's video player using embedded YouTube video.
Browser session
yt-remote-fast-check-period
Saves the settings of the user's video player using embedded YouTube video.
Browser session
yt-remote-cast-installed
Saves the user settings when retrieving a YouTube video integrated on other web pages
Browser session
yt-remote-cast-available
Saves user settings when retrieving integrated YouTube videos.
Browser session
ANID
Used for targeting purposes to profile the interests of website visitors in order to display relevant and personalized Google advertising.
2 years
SNID
Google Maps - Google uses these cookies to store user preferences and information when you view pages with Google Maps.
1 month
SSID
Used to store information about how you use the site and what advertisements you saw before visiting this site, and to customize advertising on Google resources by remembering your recent searches, your previous interactions with an advertiser's ads or search results, and your visits to an advertiser's site.
6 months
1P_JAR
This cookie is used to support Google's advertising services.
1 month
SAPISID
Used for targeting purposes to profile the interests of website visitors in order to display relevant and personalized Google advertising.
2 years
APISID
Used for targeting purposes to profile the interests of website visitors in order to display relevant and personalized Google advertising.
6 months
HSID
Includes encrypted entries of your Google account and last login time to protect against attacks and data theft from form entries.
2 years
SID
Used for security purposes to store digitally signed and encrypted records of a user's Google Account ID and last login time, enabling Google to authenticate users, prevent fraudulent use of login credentials, and protect user data from unauthorized parties. This may also be used for targeting purposes to display relevant and personalized advertising content.
6 months
SIDCC
This cookie stores information about user settings and information for Google Maps.
3 months
NID
The NID cookie contains a unique ID that Google uses to store your preferences and other information.
6 months
CONSENT
This cookie tracks how you use a website to show you advertisements that may be of interest to you.
18 years
__Secure-3PAPISID
This cookie is used to support Google's advertising services.
2 years
__Secure-3PSID
This cookie is used to support Google's advertising services.
6 months
__Secure-3PSIDCC
This cookie is used to support Google's advertising services.
6 months