The project focuses on advancing the abilities of robotic systems in performing manipulation tasks by mimicking the versatility and efficiency of human hands in manipulating objects. Without serious mental efforts, humans can readily use various tools like screwdrivers, chopsticks, hammers, pencils, brushes etc., in order to achieve a desired purpose. On the fly, they can make complex plans for transporting and manipulating objects, which require sequences of grasping, re-grasping, re-orienting, transporting, matching, pulling, pushing etc. These are common steps when performing typical service actions like (un-)zipping a rucksack, eating soup using a spoon, tying laces etc. Furthermore, human manipulation abilities are indispensable in developing skills for machining-, drilling-, polishing- and assembly operations in workshops, commonly done using just human labor. In any of these service and industrial examples, the success in performing a manipulation assignment primarily relies on the compliance of the human hand and on the ability of humans to control an object admitting its sliding or rolling over the palms or fingers.
Developing robotic hands for handling similar objects in similar settings requires new fundamentals and new scalable principles for reconstructing and re-engineering alternative solutions to the tasks. Indeed, apparent differences in kinematic structures, dynamical features, measurement systems, actuation for robots versus humans give little hope to mimic human manipulation behaviors exactly by robotic systems. Rather than that, human performances indicate a feasibility of various agile manipulation assignments done under different contact conditions and provide strong incentives for planning and controlling similar behaviors by robotic systems.
The project activities were carried out along the next conceptual directions in developing model-based solutions for a range of robotic manipulation assignments. Namely, we were searching for
1) different representation formats of manipulation primitives consistent with physical constraints, e.g. due to passive dynamics of an external object;
2) control system architectures for robust stabilization of agile movements of the system consistent with known kinematic and dynamic constraints and sets of heuristics describing a formalization of expected system performance.
The main theoretical results of the project were the following:
1. Rigorous analysis of the most demanding of singular cases of the so-called nested representation of a forced motion of a mechanical system was suggested. It explained that the common assumption imposed on the dynamics of a motion generator to be an ordinary differential equation can be alleviated. As shown, its dynamics might possess a singularity of the function that represents a factor of the highest derivative of the motion generator, which cannot be removed by algebraic operations. If such singularity is of the first order, then the contribution characterizes the properties of coefficients of the dynamic system that ensures the existence of a bounded behavior of a motion generator. Meanwhile, if such conditions are violated, or the singularity is of higher order, then bounded solutions are proved to be absent. The result is constructive, easy to check and allows both characterizing the usefulness of different choices of motion generators for performing a concrete manipulation assignment and describing extended set of manipulations if the conditions of such tests are met.
2. New control architectures for robust orbital stabilization or contraction to a nominal manipulation for underactuated robotic systems were developed. The novel element in such controller designs was the analytic procedure for reconstructing a low dimensional manifold and a feedforward transformation that makes this manifold invariant for the closed loop system. In this way, the new method admitted the use of classical high-gain and sliding mode feedback control designs for robust orbital stabilization provided that the dynamics on the induced by feedback invariant submanifold is stable.
Main outcomes: theoretical contributions
Theoretical results can be grouped around the following directions
- Developing model-based algorithms and their realizations for searching and efficient representations of feasible forced movements of underactuated mechanical systems that account for most demanding part of the dynamics of the robotic system designed for performing a dynamic manipulation with an external object or environment.
- Developing tools for model-based representation of transverse dynamics of nonlinear system in a vicinity of the nominal movement; they include steps defining different sets of transverse coordinates, deriving their dynamics and mostly important deriving symbolically the first order approximation of their dynamics.
- Developing algorithms for synthesis of a feedback controller for regulating the transverse dynamics of the nominal motion of the robotic system and for analysis of the closed-loop system dynamics.
The project will result in developing the framework and critical components of novel technology for solving common challenging robotics problems in grasping, manipulation, transporting, machining, and assembly tasks. The fundamental contributions will allow organizing original experiments and steps in prototyping scalable approaches for difficult to perform and automate operations in manufacturing and service applications. The project activities will propel the robotics research of the team to the forefront in dynamic manipulation, force control and contact applications resulting at publication of a number of reports and papers at the best robotics conference and prestige scientific journals. Organizationally, the project will serve as an integrated research, educational and experimental platform for developing new technologies, educating and training young researchers, as well as designing and testing innovative robotic solutions all necessary for substitution and automation of manual work and advanced human capabilities in sorting, high-precision manipulation, processing and assembly operations on digital factories of the future. In service applications the contributions will extend and support human-like and bio-inspired collaborative work between humans and light, flexible and intelligent robotic systems.