Systems and methods for robust robotic manipulation using chance-constrained optimization - Patents.com

Abstract

A robotic system for manipulating an object with a robotic manipulator is provided. The robotic system is configured to collect a digital representation of a task for manipulating the object and to solve a robust control problem to optimize a sequence of control forces applied by the robotic manipulator to change a state of the object. The evolution of the state of the object is governed by a stochastic complementarity system that models the task with predefined probabilities. The robust control problem optimizes a cost function to generate a sequence of control forces that executes the task subject to joint chance constraints including a first chance constraint on the state of the object being manipulated and a second chance constraint on the stochastic complementarity constraint that models the manipulation of the object. The robotic system is further configured to control the manipulation of the object based on the sequence of control forces.

Description

[Technical field]

[0001] The present invention relates generally to robotic manipulation, and more particularly to a method for robust control of robotic manipulation using chance-constrained optimization with manipulation model and parameter uncertainties. [Background technology]

[0002] Contacts play a central role in all robotic manipulation tasks. While selective use of contacts can enable robots to reason about and manipulate their environments, contact-based reasoning and control tends to be challenging in all respects. Contact models are challenging from analytical, algorithmic, and computational perspectives. As a result, little progress has been made in principled techniques for model-based manipulation. The hybrid dynamics underlying frictional interactions and uncertainties associated with friction parameters make it difficult to efficiently design model-based controllers for manipulation. Summary of the Invention [Problem to be solved by the invention]

[0003] Contact modeling has been an active area in robotics research over the last few decades. One of the most common approaches to model contact dynamics is using the linear complementarity problem (LCP). LCP models have been widely used to model contact dynamics in academia and in several physics simulation engines such as Bullet, ODE, and Havok. LCP-based contact models have been widely used to solve trajectory optimization problems in manipulation and legged locomotion. These works assume a deterministic contact model to perform the trajectory optimization. However, most friction interaction systems suffer from several sources that result in stochastic dynamics. For this reason, it is important to consider uncertainties during trajectory optimization. Uncertainties in LCP-based contact models result in stochastic complementarity constraints during optimization. Stochastic complementarity constraints tend to result in infeasibility with respect to the underlying optimization problem.

[0004] Therefore, what is needed is a system and method for solving trajectory optimization problems by considering probabilistic complementarity constraints during optimization.

[0005] It is an objective of some embodiments to provide a robotic system configured to optimize a trajectory of movement of an object being manipulated by the robotic system by optimizing a sequence of control forces acting on the object during manipulation of the object. [Means for solving the problem]

[0006] Some embodiments are based on the recognition that frictional interaction systems are subject to several uncertainties that result in stochastic dynamics. For this reason, it is important to take the uncertainties into account during trajectory optimization (TO) using a contact modeling of the frictional interaction system. Modeling the uncertainties in the LCP-based contact model results in a Stochastic Discrete-time Complementarity System (SDLCS).

[0007] Some embodiments are based on the recognition that during optimization of the SDLCS, the effect of uncertainty on the evolution of the system state should be considered in order to perform robust optimization of the system trajectory. Without such a method, the robot manipulator of the robotic system manipulating the object state may perform poorly and the resulting robotic manipulation system may fail. Currently, there are no known techniques capable of performing robust optimization for dynamic models with contact dynamics modeled as a linear complementarity system. This makes optimization and control of robotic manipulation uncertain and challenging.

[0008] Therefore, this disclosure proposes a chance-constrained optimization formulation of SDLCS for robust trajectory optimization during maneuvers. A chance-constrained Mixed-Integer Quadratic Programming (MIQPCC) is formulated to solve the optimization problem. This formulation takes into account not only joint chance constraints on complementarity, but also states to capture the stochastic evolution of the dynamics.

[0009] In some embodiments, a Stochastic Non-Linear Model Predictive Controller (SNMPC) based on the proposed formulation is designed with complementarity constraints for robotic systems such as a planar push system. SNMPC is a nonlinear model predictive controller for manipulative robotic systems.

[0010] Some embodiments are based on the recognition that robust trajectory optimization requires imposing chance constraints as hard constraints in the SDLCS. However, the hard constraints in the SDLCS make it difficult to solve the robust trajectory optimization problem. Therefore, one approach is to convert the hard constraints to soft constraints. The hard constraints can be converted to soft constraints by including them in a cost function that is optimized to obtain a robust trajectory for maneuvering an object. Furthermore, the SDLCS with soft constraints undergoes expected residual minimization (ERM)-based penalty to obtain an optimized trajectory by solving the trajectory optimization problem. However, the optimized trajectory obtained with the ERM-based penalty and soft constraints for the SDLCS is not very robust because the formulation of this approach does not consider the stochastic state evolution of the system during optimization.

[0011] Therefore, the robotic system of the present disclosure solves the trajectory optimization problem by imposing the chance constraint as a hard constraint on the SDLCS. For this, the chance constraint is split into two modes by relaxing the chance constraint so that the violation of the chance constraint falls within an acceptable range. In this way, the SDLCS with the hard chance constraint is optimized to obtain a robust trajectory for manipulating an object.

[0012] Accordingly, one embodiment discloses a robotic system including a robotic manipulator, a processor, and a memory having instructions stored thereon that, when executed by the processor, cause the robotic manipulator to perform the following operations: collecting a digital representation of a task for manipulating an object from an initial state to a final state. The processor is further configured to solve a robust control problem to optimize a sequence of control forces applied by the robotic manipulator to change a state of the object from the initial state to the final state, the evolution of the state of the object being governed by a stochastic complementarity system that models the task of manipulating the object with a predefined probability. The robust control problem optimizes a cost function to generate the sequence of control forces that perform the task subject to joint chance constraints including a first chance constraint on the state of the object being manipulated and a second chance constraint in the stochastic complementarity constraints that models the manipulation of the object by the robotic manipulator. The processor is further configured to control the manipulation of the object by applying the sequence of control forces that change the state of the object from the initial state to the final state.

[0013] Accordingly, another embodiment discloses a method for manipulating an object using a robotic system. The method includes collecting a digital representation of a task for manipulating an object from an initial state to a final state. The method further includes solving a robust control problem to optimize a sequence of control forces applied by the robotic manipulator to change a state of the object from the initial state to the final state, the evolution of the state of the object being governed by a stochastic complementarity system that models the task of manipulating the object with a predefined probability. The robust control problem optimizes a cost function to generate the sequence of control forces that execute the task subject to joint chance constraints including a first chance constraint on the state of the object being manipulated and a second chance constraint in the stochastic complementarity constraints that model the manipulation of the object by the robotic manipulator. The method further includes controlling the manipulation of the object by applying the calculated sequence of control forces that change the state of the object from the initial state to the final state.

[0014] DETAILED DESCRIPTION OF THE DRAWINGS

[0023] Embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. The drawings shown are not necessarily to scale, with the emphasis generally being placed on illustrating the principles of embodiments of the present disclosure. [Brief description of the drawings]

[0015] [Figure 1] FIG. 1 is a block diagram illustrating a stochastic discrete-time linear complementarity system (SDLCS) based robotic system for manipulating an object in accordance with an illustrative embodiment. [Diagram 2] FIG. 1 illustrates a robust trajectory (TO) optimization for SDLCS in accordance with an illustrative embodiment. [Diagram 3] FIG. 1 illustrates a method for imposing joint linear chance constraints in accordance with an illustrative embodiment. [Figure 4]FIG. 1 illustrates a decomposition of a probabilistic complementarity constraint into two modes in accordance with an exemplary embodiment. [Diagram 5] FIG. 1 illustrates a diagram illustrating the relaxation of deterministic chance complementarity constraints in probabilistic chance complementarity constraints in accordance with an exemplary embodiment. [Figure 6] FIG. 1 illustrates an objective function and stochastic dynamics for MIQPCC in accordance with an illustrative embodiment. [Figure 7] FIG. 1 is a block diagram illustrating a robotic system for manipulating an object in accordance with an illustrative embodiment. [Figure 8] FIG. 1 illustrates a manipulation system configured to push an object with a manipulator in accordance with an illustrative embodiment. [Figure 9] 1 illustrates method steps for manipulating an object using a robotic system in accordance with an exemplary embodiment. [Figure 10] FIG. 1 illustrates a robotic system configured to move an object from a work surface to a bin in accordance with an illustrative embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the present disclosure. However, it will be apparent to those skilled in the art that the present disclosure may be practiced without these specific details. In other instances, the devices and methods are shown only in block diagram form in order to avoid obscuring the present disclosure.

[0017] As used herein and in the claims, the words "for example," "for example," and "such as," as well as "comprising," "having," "including," and other verb forms thereof, when used in conjunction with a list of one or more components or other items, should be construed as open-ended. This means that the list should not be considered to exclude additional components or items. The term "based on" means based at least in part on. Furthermore, it should be understood that the terms and terminology used herein are for purposes of description and should not be considered as limiting. Any headings used herein are for convenience only and should not be construed as having any legal or limiting effect.

[0018] FIG. 1 illustrates a block diagram of a stochastic discrete-time linear complementarity system (SDLCS)-based robotic system 101 for manipulating an object 105, according to an exemplary embodiment. The robotic system 101 comprises a robotic manipulator 103, and the robotic system 101 is configured to control the robotic manipulator 103 to manipulate the object 105. The manipulation of the object 105 may correspond to sliding the object 105 from an initial state (or position) 107a to a final state (or position) 107b along one or more trajectories 109. To that end, the robotic system 101 is configured to determine a sequence 111 of control forces to be applied to the object 105 via the robotic manipulator 103 to move the object 105 from the initial state 107a to the final state 107b along one or more trajectories 109. In some embodiments, the manipulation of the object 105 may correspond to gripping the object 105 at the initial state 107a and moving it to the final state 107b.

[0019] To efficiently manipulate the object 105, it is important to realize an efficient contact model that includes the interaction between the contact surfaces, i.e., between the robot manipulator 103 and the object 105. The contact model includes trajectory optimization for manipulating the object 105.

[0020] Some embodiments are based on the recognition that a friction interaction system, such as a robotic system 101 configured to manipulate an object, e.g., slide an object 105, is subject to several uncertainties that may lead to stochastic dynamics. It is therefore important to take the uncertainties into account during TO.

[0021] To that end, an SDLCS with parameter uncertainties and additive noise in dynamics and complementarity constraints is realized to interface model the robotic system 101. As shown in Fig. 1, the uncertainties lead to stochastic evolution of the system states in the SDLCS. For this reason, the robust optimization formulation must consider the SDLCS uncertainties in the state evolution.

[0022] The SDLCS-based robotic system 101 is configured to collect a digital representation of a task for manipulating an object 105 from an initial state 107a to a final state 107b. The SDLCS manages the evolution of the task for manipulating the object 105 from the initial state 107a to the final state 107b with a predefined probability. The SDLCS includes the interaction uncertainty of the robotic manipulator 103 with respect to the object 105. The interaction uncertainty in the SDLCS is introduced by one or a combination of the following: noise in the evolution of the state of the object 105, noise in the measurement of the state of the object 105, uncertainty in the model of the robotic manipulator 103, uncertainty in the environment of the robotic manipulator 103, etc. In an exemplary embodiment in which the robotic manipulator 103 is configured to slide / push the object 105 on a surface, the interaction uncertainty in the SDLCS is introduced by uncertainty in the friction coefficient of the surface.

[0023] The robotic system 101 is further configured to solve a robust control problem to optimize a sequence 111 of control forces applied by the robotic manipulator 103 to modify the state of the object 105, the robust control problem optimizing a cost function to generate a sequence 111 of control forces that executes the task of manipulating the object 105 subject to chance constraints. A chance constrained method is used to solve the trajectory optimization problem under various uncertainties. It is a formulation of the optimization problem that ensures that the probability of satisfying certain constraints is above a certain level. In other words, it restricts the feasible region such that the confidence level of the solution is high. The chance constrained method is a relatively robust approach.

[0024] To that end, the robotic system 101 is configured to solve a robust control problem with joint chance constraints, the joint chance constraints including a first chance constraint on a state of the object 105 being manipulated and a second chance constraint on a probabilistic complementarity constraint that models the manipulation of the object by the robotic manipulator 103. That is, the probabilistic complementarity constraints include the uncertainty of the interaction of the robotic manipulator 103 with the object 105 being manipulated. Furthermore, the robotic system 101 is configured to calculate a sequence of control forces 111 by solving the robust control problem with the joint chance constraints, and to control the manipulation of the object 105 by applying the calculated sequence of control forces 111 that changes the state of the object 105 from an initial state 107a to a final state 107b.

[0025] In some embodiments, the control of the manipulation of the object 105 is a feedforward control according to a sequence of control forces 111 that changes the state of the object 105 from the initial state 107a to the final state 107b. In some embodiments, the control is a feedback control that updates the sequence of control forces 111 that changes the state of the object 105 from the initial state 107a to the final state 107b in response to receiving a feedback signal indicative of the current state of the object 105. The feedback control is a predictive control that iteratively optimizes the sequence of control forces 111 over a prediction horizon subject to the hard constraints of the stochastic complementarity system. Mathematical implementation:

[0026] SDLCS is based on the Discrete-Time Linear Complementarity System (DLCS), the details of which are described below. I. Discrete-Time Linear Complementary Systems (DLCS)

[0027] A DLCS is a discrete-time dynamical system whose state evolution is governed by the linear dynamics of the states and algebraic variables that solve the LCP. A DLCS with complementarity constraints is given by:

number

[0028] The SDLCS, i.e., DLCS with uncertainty, is given by:

number

[0029] A.Robust trajectory optimization for SDLCS

[0030] 2 illustrates a method for robust TO for SDLCS in accordance with an example embodiment. A formulation for robust TO of SDLCS is given by objective function 201.

number

[0031] Some embodiments are based on the following assumptions about equations (5)-(8):

number

[0032] B. Joint Linear Chance Constraints

[0033]

number

[0034] Some embodiments are based on the recognition that it is difficult to obtain the cumulative distribution function (cdf) of equation (9). Therefore, in step 303, Boolean inequality is employed to obtain a conservative approximation of (9) as follows:

number

[0035]

number

[0036] C. Chance Complementarity Constraint (CCC) for SDLCS

[0037]

number

[0038] Some embodiments are based on the recognition that robust trajectory optimization requires imposing chance constraints as hard chance constraints in SDLCS. However, using hard constraints in SDLCS makes the robust trajectory optimization problem difficult to solve. Therefore, the stochastic complementarity constraints are decomposed into two modes so that the stochastic complementarity constraints are mathematically easier or tractable.

[0039] 4 illustrates a decomposition 401 of a probabilistic complementarity constraint into two modes 403 and 405 according to an exemplary embodiment. The probabilistic complementarity constraint is decomposed into two disjoint inequality modes. The two modes are a first mode 403 (also referred to as mode 1) and a second mode 405 (in equation (16)) (also referred to as mode 2) after the decomposition 401 of the probabilistic complementarity constraint, as follows:

number

[0040]

number

[0041]

number

[0042]

number

[0043]

number

[0044]

number

[0045] D. Chance-constrained mixed integer quadratic programming

[0046] Some embodiments are based on the recognition that since the chance constraints (equations (5)-(8)) contain quadratic objective terms, the trajectory optimization problem (i.e., the robust control problem) is solved by chance-constrained mixed integer quadratic programming (MIQPCC). Figure 6 shows the objective function and stochastic dynamics evolution for MIQPCC according to an example embodiment. The MIQPCC formulation for solving equations (5)-(8) may be given as follows:

number

[0047]

number

[0048] In some embodiments, the CCC is used to design a Stochastic Non-linear Model Predictive Control (SNMPC) for a manipulation system. A formulation for implementing SNMPC for a Stochastic Nonlinear Complementarity System (SNCS) is described below.

[0049] E. Stochastic Nonlinear Model Predictive Control (SNMPC)

number

[0050]

number

[0051] The modified chance constrained mixed integer quadratic programming (MIQPCC) for SNMPC is given by:

number

[0052]

number

[0053] 7 shows a block diagram 700 of a robotic system 101 for manipulating an object 105 according to an exemplary embodiment. The robotic system 101 may have several interfaces that connect the robotic system 101 with other systems and devices. For example, a network interface controller (NIC) 701 is adapted to connect the robotic system 101 to a network 705 via a bus 703. The robotic system 101 may receive a digital representation 707 of one or more tasks through the network 705, wirelessly, or by wire. The one or more tasks 707 include at least one of sliding the object 105 from an initial position to a final position or gripping and moving the object 105 from an initial position to a final position.

[0054] The robotic system 101 includes a processor 711 configured to execute stored instructions. The robotic system 101 further includes a memory 713 that stores instructions executable by the processor 711. The processor 711 may be a single-core processor, a multi-core processor, a computing cluster, or any number of other configurations. The memory 713 may include random access memory (RAM), read only memory (ROM), flash memory, or any other suitable memory system. The processor 711 is connected to one or more input / output devices via a bus 703. Additionally, the robotic system 101 includes a storage device 715 adapted to store various modules that store executable instructions for the processor 711. The storage device 715 may be implemented using a hard drive, an optical drive, a thumb drive, an array of drives, or any combination thereof.

[0055] The storage device 715 is configured to store the SDLCS 715a and the chance complementarity constraints 715b. Upon receiving the digital representation 707 of one or more tasks, the processor 711 is configured to solve a robust control problem to optimize a sequence of control forces applied by the robotic system 101 to manipulate an object, the evolution of the state of the object being governed by the SDLCS 715a that models the task of manipulating the object with a predefined probability. The processor 711 is further configured to optimize a cost function to generate a sequence of control forces that execute the task subject to the chance complementarity constraints 715b, the chance complementarity constraints including a first chance constraint on the state of the object being manipulated and a second chance constraint on the probabilistic complementarity constraint that models the manipulation of the object. The processor 711 is further configured to control the manipulation of the object based on the calculated sequence of control forces.

[0056] Additionally, the robotic system 101 may include an output interface 717. In some embodiments, the robotic system 101 is further configured to send the sequence of control forces via the output interface 717 to a controller 719 configured to control a robotic manipulator to manipulate an object based on the sequence of control forces. In some embodiments, the robotic manipulator may be directly controlled by the processor 711 based on the calculated sequence of control forces.

[0057] FIG. 8 illustrates a manipulation system 801 configured to push an object 805 with a manipulator 803 according to an exemplary embodiment. The manipulation system 801 is based on SDLCS, where the manipulation system 801 is a pusher-slider system. Furthermore, the manipulator 803 may correspond to a pusher and a slider. FIG. 8 illustrates two possible sources of uncertainty in the manipulation system 801, namely the uncertain friction cone and the contact point. The dynamics of the manipulation system 801 are given by:

number

[0058]

number

[0059]

number

[0060] 9 illustrates steps of a method 900 for manipulating an object using the robotic system 101, according to an exemplary embodiment. FIG. 9 is described below in conjunction with FIG.

[0061] In step 901, a digital representation of a task, such as manipulating an object 105 from an initial position to a final position, is collected, where the object 105 can be manipulated using a robotic manipulator 103 of the robotic system 101. The robotic manipulator 103 applies a sequence of control forces 111 to the object 105 to manipulate the object 105. The robotic manipulator 103 is at least one of a robot wrist, a gantry robot, a cylindrical robot, a polar robot, and a jointed arm robot with an end effector, such as a parallel jaw gripper. Additionally, the robotic system 101 employs a SDLCS, which manages the evolution of a task to manipulate the object 105 from an initial state 107a to a final state 107b with predefined probabilities.

[0062] In step 903, the method 900 includes solving a robust control problem to optimize a sequence of control forces 111 to be applied by the robotic manipulator 103 to change the state of the object 105. The robust control problem optimizes a cost function to generate a sequence of control forces 111 that accomplishes the task of manipulating the object 105 subject to chance constraints. A chance constrained method is used to solve the trajectory optimization problem under various uncertainties.

[0063] To that end, the method 900 includes solving a robust control problem with joint chance constraints, the joint chance constraints including a first chance constraint on a state of the object 105 being manipulated and a second chance constraint on a probabilistic complementarity constraint that models the manipulation of the object by the robotic manipulator 103. To that end, the method 900 includes calculating the sequence of control forces 111 by solving the robust control problem with the joint chance constraints.

[0064] At step 905, the method 900 includes controlling the manipulation of the object 105 by applying the calculated control forces to change the state of the object 105 from the initial state 107a to the final state 107b.

[0065] FIG. 10 illustrates a robotic system 1000 configured to move an object 1001 from a work surface 1003 to a bin 1005 according to an exemplary embodiment. In this description, the robotic system 1000 is a set of components 1007, 1009, 1011, 1013 connected by joints 1015, 1017, 1019, 1021, 1023. In the described embodiment, the joints 1015, 1017, 1019, 1021, 1023 are revolute joints, but in alternative embodiments, they may be sliding joints or other types of joints. The collection of joints determines the degrees of freedom for a robotic arm 1027. The robotic arm 1027 has five degrees of freedom for each of the joints 1015, 1017, 1019, 1021, 1023. In alternative embodiments, the robot may include six joints. An end effector 1025 is attached to the robotic arm 1027. An end effector 1025 is attached to one of the components, typically the last component 1013 if the components are determined to be in a chain. The end effector 1025 may be a parallel jaw gripper. A parallel jaw gripper has two parallel fingers whose distance can be adjusted relative to one another.

[0066] Many other end effectors can be used instead, including, for example, welding tips. The joints 1015, 1017, 1019, 1021, 1023 can be adjusted to achieve a desired configuration for the component. The desired configuration may relate to a desired position in Euclidean space, or a desired value in joint space. The joints are also controllable in the time domain to achieve a desired (angular) velocity and / or (angular) acceleration. The joints have embedded sensors that can report the state of the joint. The reported state can be angle, current, velocity, torque, acceleration, or any combination thereof. The collection of reported joint states is referred to as the robot state 1029. The robot state 1029 may be used for chance constraints in probabilistic complementarity constraint modeling manipulation of the object 1001 by the robot arm 1027.

[0067] Commands for the robot arm 1027 are received from the robot controller 1031 via connection 1033, and the robot state 1029 is received by the robot controller 1031 via connection 1033. In a preferred embodiment, connection 1033 is a dedicated data cable. In another embodiment, connection 1033 is an Ethernet cable. The robot controller 1031 may be configured to perform a variety of different tasks. For example, the robot controller 1031 may be configured to control the robot arm 1027 to pick an object 1001 from a work surface 1003 and place it in a bin 1005, which may be configured to collect a number of objects.

[0068] The robot controller 1031 moves the object 1001 from the working surface 1003 to the bin 1005 using the proposed SDLCS and chance complementarity constraints for robust trajectory optimization. To do so, the SDLCS-based robot controller 1031 first collects a digital representation of the task of grasping the object 1001 from the working surface 1003. The position and orientation of the working surface 1003 and the object 1001 may be part of the digital representation of the task. The robot controller 1031 is further configured to solve a robust control problem to optimize a sequence of control forces applied by the robot controller 1031 to the robot arm 1027 to grasp the object 1001 from the working surface 1003, where the robust control problem optimizes a cost function to generate a sequence of control forces that executes the task of grasping the object 1001 subject to chance constraints.

[0069] The robot controller 1031 is further configured to solve the robust control problem using the joint chance constraints, the joint chance constraints including a first chance constraint on a state of the object 1001 being manipulated and a second chance constraint on a probabilistic complementarity constraint that models the manipulation of the object by the robot arm 1027. That is, the probabilistic complementarity constraints include an uncertainty of an interaction of the robot arm 1027 with the object 1001 being manipulated. The second chance constraint may be based on the robot state 1029. Furthermore, the robot controller 1031 is further configured to calculate a sequence of control forces by solving the robust control problem using the joint chance constraints, and control the manipulation of the object 1001 by applying the calculated sequence of control forces to grasp and move the object 1001 from the work surface 1003 to the bin 1005.

[0070] The various methods or processes outlined herein may be coded as software executable on one or more processors employing any one of a variety of operating systems or platforms. In addition, such software may be written using any of a number of suitable programming languages ​​and / or programming or scripting tools, and compiled as executable machine language code or intermediate code that runs on a framework or virtual machine. Typically, the functionality of the program modules may be combined or distributed as desired in various embodiments.

[0071] Also, embodiments of the invention may be embodied as methods, examples of which are provided. The acts performed as part of the method may be ordered in any suitable manner. Thus, while illustrated as sequential operations in exemplary embodiments, embodiments may be constructed in which operations are performed in different orders than as illustrated, and may include performing some operations simultaneously. Furthermore, the use of ordinal terms such as "first," "second," and the like in the claims to modify elements of the claims does not, in itself, imply any priority, precedence, or order of the elements of one claim relative to elements of another claim, or any chronological order in which the operations of the method are performed, but is merely used as a label to distinguish elements of one claim having a certain name from other elements having the same name (except when ordinal terms are used) to distinguish between elements of the claims.

[0072] Although the present disclosure has been described with reference to certain preferred embodiments, it should be understood that various other adaptations and modifications can be made within the spirit and scope of the present disclosure. It is therefore intended in the scope of the appended claims to cover all such variations and modifications as fall within the true spirit and scope of the present disclosure.

Claims

1. 1. A robot system comprising: A robot manipulator; A processor; and a memory storing instructions that, when executed by the processor, cause the robotic manipulator to perform the following operations: collecting a digital representation of a task for manipulating an object from an initial state to a final state; and solving a robust control problem to optimize a sequence of control forces applied by the robotic manipulator to change a state of the object from the initial state to the final state, wherein the evolution of the state of the object is governed by a stochastic complementarity system that models the task of manipulating the object with a predefined probability, the robust control problem optimizing a cost function to generate the sequence of control forces that execute the task subject to joint chance constraints including a first chance constraint on the state of the object being manipulated and a second chance constraint in a stochastic complementarity system that models the manipulation of the object by the robotic manipulator, and the following operations further include: controlling the manipulation of the object by applying the sequence of control forces that change the state of the object from the initial state to the final state.

2. 2. The robotic system of claim 1, wherein the probabilistic complementarity system is a probabilistic discrete-time linear complementarity system (SDLCS) that includes uncertainty of the robotic manipulator's interaction with the object being manipulated, and the complementarity constraints model the object's interaction together with the sequence of control forces during manipulation of the object.

3. 3. The robotic system of claim 2, wherein the uncertainty of the interaction in the stochastic discrete-time linear complementarity system (SDLCS) is caused by one or a combination of noise in the evolution of the state of the object, noise in the measurement of the state of the object, uncertainty in a model of the robot manipulator, and uncertainty in an environment of the robot manipulator.

4. 3. The robotic system of claim 2, wherein the robotic manipulator manipulates the object placed on a surface, and the uncertainty of the interaction in the stochastic discrete-time linear complementarity system (SDLCS) is caused by the uncertainty in a coefficient of friction of the surface.

5. 3. The robotic system of claim 2, wherein the robotic manipulator is at least one of a robot wrist, a gantry robot, a cylindrical robot, a polar coordinate robot, and a jointed arm robot with an end effector such as a parallel jaw gripper.

6. The robotic system of claim 1 , wherein the probabilistic complementarity constraints are relaxed to allow violation of the probabilistic complementarity constraints with a probability defined by a hyper-parameter.

7. 7. The robotic system of claim 6, wherein the hyperparameters are selected as a function of predefined probabilities such that the probabilistic complementarity system is satisfied with a predefined probability as a result of achieving a relaxed complementarity constraint with the probability defined by the hyperparameters.

8. 7. The robotic system of claim 6, wherein the probabilistic complementarity system includes two complementarity variables, and the relaxation is imposed on individual probabilities for one or both of the complementarity variables to satisfy an inequality of the complementarity constraint.

9. 9. The robotic system of claim 8, wherein a first complementary variable is deterministic and a second complementary variable is probabilistic with a relaxed satisfaction of an inequality that depends on the deterministic value of the first complementary variable.

10. The robotic system of claim 6 , wherein the probabilistic complementarity constraints are decomposed into disjoint inequality modes, and the relaxation imposes chance constraints on the occurrence of each mode.

11. The robotic system of claim 1 , wherein the robust control problem is solved by a chance-constrained mixed integer quadratic programming method.

12. The robotic system of claim 1 , wherein the control is a feedforward control according to a sequence of the control forces that change the state of the object from the initial state to the final state.

13. 2. The robotic system of claim 1, wherein the control is a feedback control that updates the sequence of control forces that change the state of the object from the initial state to the final state in response to receiving a feedback signal indicative of a current state of the object.

14. The robotic system of claim 13 , wherein the feedback control is a predictive control that iteratively optimizes the sequence of control forces over a predictive range subject to hard constraints of the stochastic complementarity system.

15. 1. A method for manipulating an object with a robotic system, comprising: collecting a digital representation of a task for manipulating an object from an initial state to a final state; and solving a robust control problem to optimize a sequence of control forces applied by a robotic manipulator to change a state of the object from the initial state to the final state, wherein the evolution of the state of the object is governed by a stochastic complementarity system that models the task of manipulating the object with predefined probabilities, the robust control problem optimizing a cost function to generate the sequence of control forces that execute the task subject to joint chance constraints including a first chance constraint on the state of the object being manipulated and a second chance constraint in a stochastic complementarity system that models the manipulation of the object by the robotic manipulator, the method further comprising:

11. A method comprising: controlling the manipulation of the object by applying a sequence of the control forces that change the state of the object from the initial state to the final state.

16. 16. The method of claim 15, wherein the stochastic complementarity system is a stochastic discrete-time linear complementarity system (SDLCS) that includes uncertainty of the interaction of the robotic manipulator with the object being manipulated, and the complementarity constraints model the interaction of the object together with the sequence of control forces during manipulation of the object.

17. 17. The method of claim 16, wherein the uncertainty of the interaction in the stochastic discrete-time linear complementarity system (SDLCS) is caused by one or a combination of noise in the evolution of the state of the object, noise in the measurement of the state of the object, uncertainty in a model of the robot manipulator, and uncertainty in an environment of the robot manipulator.

18. 17. The method of claim 16, wherein the robotic manipulator manipulates the object placed on a surface, and the uncertainty of the interaction in the stochastic discrete-time linear complementarity system (SDLCS) is caused by the uncertainty in a coefficient of friction of the surface.

19. 17. The method of claim 16, wherein the robotic manipulator is at least one of a robot wrist, a gantry robot, a cylindrical robot, a polar robot, and a jointed arm robot with an end effector such as a parallel jaw gripper.

20. The method of claim 15 , wherein the probabilistic complementarity constraints are relaxed to allow violation of the probabilistic complementarity constraints with a probability defined by a hyperparameter.

Images