18481164. LEARNING LATENT-SPACE BARRIER FUNCTIONS FROM SAFE DEMONSTRATIONS simplified abstract (TOYOTA JIDOSHA KABUSHIKI KAISHA)

LEARNING LATENT-SPACE BARRIER FUNCTIONS FROM SAFE DEMONSTRATIONS

Organization Name

TOYOTA JIDOSHA KABUSHIKI KAISHA

Inventor(s)

Fernando Castaneda Garcia-rozas of Berkeley CA (US)

Haruki Nishimura of Sunnydale CA (US)

Rowan Mcallister of San Jose CA (US)

Adrien David Gaidon of Mountain View CA (US)

LEARNING LATENT-SPACE BARRIER FUNCTIONS FROM SAFE DEMONSTRATIONS - A simplified explanation of the abstract

This abstract first appeared for US patent application 18481164 titled 'LEARNING LATENT-SPACE BARRIER FUNCTIONS FROM SAFE DEMONSTRATIONS

The method described in the patent application is for a task-agnostic policy filter control system. It involves encoding a current observation, a previous latent state, and a previous action to output a new latent state. This new latent state is then used to compute learned latent state-space dynamic models through a neural ordinary differential equations (ODE) module. An in-distribution barrier function (iDBF) model is used to infer an iDBF value in response to the new latent state. Based on the learned dynamic models, the iDBF value, and a reference control input, a current action is computed to keep the system in-distribution with respect to a dataset of safe demonstrations.

• Encoding current observation, previous latent state, and previous action
• Computing learned latent state-space dynamic models using a neural ODE module
• Inferring iDBF value with an in-distribution barrier function model
• Computing current action based on learned models, iDBF value, and reference control input
• Keeping the system in-distribution with respect to a dataset of safe demonstrations

Potential Applications: - Autonomous vehicles - Robotics - Industrial automation - Healthcare systems - Surveillance systems

Problems Solved: - Ensuring safety in autonomous systems - Improving control and decision-making processes - Enhancing system adaptability and responsiveness

Benefits: - Increased safety in autonomous operations - Enhanced system performance and efficiency - Improved decision-making capabilities - Adaptability to changing environments

Commercial Applications: Title: "Enhancing Safety and Performance in Autonomous Systems with Task-Agnostic Policy Filter Control" This technology can be applied in various industries such as autonomous vehicles, robotics, industrial automation, healthcare systems, and surveillance systems. It can improve safety, performance, and decision-making processes in these applications, leading to increased efficiency and reliability.

Questions about the technology: 1. How does the task-agnostic policy filter control system ensure safety in autonomous operations? 2. What are the key advantages of using neural ODE modules in computing learned dynamic models for the system?

Original Abstract Submitted

A method for a task-agnostic policy filter control system is described. The method includes encoding a current observation, a previous latent state, and a previous action to output a new latent state. The method also includes computing, by a neural ordinary differential equations (ODE) module, learned latent state-space dynamic models for the new latent state. The method further includes inferring, by an in-distribution barrier function (iDBF) model, an iDBF value in response to the new latent state. The method also includes computing, based on the learned latent state-space dynamic models, the iDBF value and a reference control input for a current timestep, and a current action. The current action keeps the task-agnostic policy filter control system in-distribution with respect to an offline-collected dataset of safe demonstrations.