IPAB Workshop - 24/10/19

 

 

Vladimir Ivan

 

Title: Loco-manipulation using non-linear programming formulations for industrial applications Abstract: Industrial applications hint on flexibility and agility alongside productivity, particularly in small and medium enterprises. How does our research address these requirements? In this talk, I'll introduce our integration work on a high-performance omni-directional mobile manipulation platform with integrated whole-body control, real-time collision-free whole-body motion planning, and perception. While these ideas are not novel to the academic community, their integration results a system that can be programmed in a flexible way to implement a variety of manipulation and locomotion tasks. I will talk about practical issue we overcome and lessons learned when using these formulations do build demonstrators for industry driven applications in asset inspection and maintenance.

 

Guiyang Xin

 

Title: The comparison between semi-analytical and fully optimization-based whole-body controllers for legged robots

Summary: Controller is an essential modul in a robotic system. Since legged robots rely on contact constraints with the environments,  controllers for them should track desired trajectories accurately and guarantee physical feasibility as well. Optimization-based controllers for torque controllable legged robots have been developed to deal with inequality physical constraints in recent years. Fully optimization-based controllers totally rely on optimization solver to compute torque commands that satisfy multiple tasks and constraints. In contrast, a semi-analytical whole-body controller is composed of task-space inverse dynamics and a quadratic programming problem subject to inequality constraints. The semi-analytical controller has advantages in terms of computation time and extensible applications. In this talk, I will compare the controllers that have been applied on ANYmal, and share my practical experience.

 

Yordan Hristov

 

Title: Disentangled Relational Representations for Explaining and Learning from Demonstration Abstract: Learning from demonstration is an effective method for human users to instruct desired robot behaviour. However, for most non-trivial tasks of practical interest, efficient learning from demonstration depends crucially on inductive bias in the chosen structure for rewards/costs and policies. We address the case where this inductive bias comes from an exchange with a human user. We propose a method in which a learning agent utilizes the information bottleneck layer of a high-parameter variational neural model, with auxiliary loss terms, in order to ground abstract concepts such as spatial relations. The concepts are referred to in natural language instructions and are manifested in the high-dimensional sensory input stream the agent receives from the world. We evaluate the properties of the latent space of the learned model in a photorealistic synthetic environment and particularly focus on examining its usability for downstream tasks. Additionally, through a series of controlled table-top manipulation experiments, we demonstrate that the learned manifold can be used to ground demonstrations as symbolic plans, which can then be executed on a PR2 robot.

 

 

Oct 24 2019 -

IPAB Workshop - 24/10/19

Vladimir Ivan, Guiyang Xin

IF, G.03