Safety standards for human-robot interaction
The original safety standard for traditional industrial robots is the ISO 10218 initially edited in 1992. During its last revision, the decision was taken to separate it into two parts: ISO 10218-1 and ISO 10218-2 lately revised in 2011 [Fryman 2014]. On the one hand, ISO 10218-1 [ISO 2011a] is particularly dedicated to manufacturers; It describes dangerous phenomena related to robots and gives recommendations and guidelines for protective measures and safe design indications in line with the machinery directive 2006/42/EC. Technical speci cations in this part concern, in a non-exhaustive way, de nitions and requirements related to: singularity, protection against hazards that can be caused by mechanical transmission, requisites in case of loss of power, performances of safety control circuits, conditions relative to control modes and limitations of power and strength plus the description of emergency stop functions. On the other hand, ISO 10218-2 is drafted to give more diverse safety requirements and guidance to integrators for the installation of industrial robots and industrial robot systems \robot+cell(s) ». ISO 10218-2 [ISO 2011b] describes how to properly conduct a task-based risk assessment to eliminate or reduce risks associated with hazards to personnel: hazards engendered by the design, implementation and use of these systems. The newest and most exciting utility introduced in ISO 10218-1 and -2 is the direct interaction between robotic systems and human-operators. Indeed, historically, for safety reasons, industrial robots are usually separated from humans by the mean of physical segregation, i.e., cages. ISO 10218-1 and -2 give general guidelines regarding safety requirements for human-robot interaction but do not provide the needed metrics for the design of collaborative robots, tasks and control strategies. In these standards, four collaborative techniques are considered:
• Safety-rated monitored stop: the system does a monitored stop so the human can enter its workspace to perform an assigned task. The robot system resumes its autonomous task once the operator is out.
• Hand-guiding: based on a safety-rated monitored stop, a human-operator can take control of the end-e ector to move the robot. Once motion is complete, a monitored stop is issued again as the person exists the workspace.
• Speed separation monitoring: the robot arm and the human maintain a safe distance from each other. Depending on this distance, the speed of the robot is reduced.
• Power and force limiting: Power and force of the robot are limited by design or control during physical interaction.
The latest document from the International Standardization Organization (ISO) fully dedicated to human-robot collaboration is the ISO/TS 15066 [ISO 2016]. Issued in 2016, this Technical Speci cation (TS) is a game changer for the robotics industry as it gives speci c and detailed metric-based safety guidance needed to assess and control risks during human-robot interaction. Pressure, force and even transferred energy limit values based on injury and pain sensitivity thresholds for various areas of the human body are provided. It states for example the following limits for allowing human-robot interaction at the hands level: maximum transferred energy E = 0:49 J and maximum permissible force F = 140 N.
Considerations when dealing with safety during human-robot interac-tion
Although the presented state-of-the-art gives various propositions on how to handle safety during human-robot interaction, there are still numerous issues to be addressed, speci cally in terms of satisfying the safety limit values recommended by the recent ISO/TS 15066 standard [ISO 2016] and the discontinuity problems that may occur when switching between control modes for dif-ferent interaction phases, e.g., at collisions and at contact establishment and release. In this sense, the work presented in this thesis aims at closing the gap and de ning a generic framework to control a robotic arm during di erent interaction phases with a human-operator. Safety in the controller is therefore included as a constraint.
The following aspects are considered for the formulation of safety constraints and a safe human-robot interaction controller:
• Safety indicators I related to the safety limit values provided in ISO/TS 15066 [ISO 2016] that can be expressed at any point of a multi-body robot must be formulated. These indicators should re ect the amount of danger the robot exhibits towards a nearby human-operator during di erent interaction phases: as the human enters the workspace of the robot, after the establishment of a deliberate or non-deliberate physical contact and as contact is released.
• The formulation of safety criteria Ilimit that represent the maximum values allowed for the safety indicators and that consequently bound the dynamics of the robot for each interaction phase. These safety criteria should also be based on the safety metrics given by the ISO/TS 15066 standard [ISO 2016].
• Finally using the safety indicators and safety criteria, the formulation of safety constraints that depend on the joint control input : I( ) Ilimit. These constraints can then easily be included in an optimization control scheme.
The di erent safety indicators from the literature are usually only useful during one interac-tion phase. Velocity of the end-e ector of the robot for example is signi cant only during the movements of the robot and therefore cannot be used as a safety indicator during physical con-tact. Contact forces on the other hand can only be evaluated during physical interaction. More generic indicators should therefore be proposed.
For a given shape of the contact surface, two parameters are source of danger during human-robot interaction: the impact force created at collision and the contact force generated after the establishment of a physical contact. The most generic way to include and consider these parameters for the formulation of safety indicators is to use energy. Indeed, energy is a univer-sal quantity that can describe most of the physical phenomena occurring during human-robot interaction. The impact force for example is directly related to the amount of kinetic energy dissipated at collision and the contact forces mostly derive from the amount of potential energy that accumulates in controller of the robot during physical contact. Velocity, inertia and also the resulting position error during physical contact are all part of the mathematical expression of energy. Safety during human-robot interaction can therefore be ensured by modulating the amount of energy instantaneously deployed by the robot. The main contributions presented in Chapter 3 regarding safety during human-robot interaction are as follows:
• the formulation of kinetic and potential energy1 based safety indicators that represent the degree of danger of a robotic arm at collision and during physical contact.
• The formulation of safety criteria that bound the dynamics of the robot during di erent interaction phases: at the approach of a human-operator, at collision, during physical contact and nally as contact is released.
• The expression of these indicators and criteria as inequality constraints related to the torque control input of the robot.
• The introduction of the concept of \task energy pro le » to track the amount of energy used to accomplish a considered repetitive task in the most optimal way. This pro le is then used to constrain the instantaneous amount of energy the robot is allowed to exhibit at every time-step. The resulting controller renders the robot compliant to any deliberate or non-deliberate contact with its environment.
For the validation of the proposed algorithms, beyond theoretical contributions, simulation re-sults conducted with a KUKA LWR4 serial robot that physically interacts with its environment are also presented in Chapter 3. Chapter 4 on the other hand presents the results of exper-imental applications on a real KUKA LWR4 whose energy is modulated as a human-operator enters its workspace, physically interacts with it, release contact then safely leave the workspace. In this scenario, the amount of kinetic energy deployed by the KUKA LWR4 during the approach of the human is modulated according to the operator-robot real-time distance. This distance is computed with an algorithm that uses point clouds acquired from the scene using a set of depth sensors (Kinects).
Considerations when dealing with constraints
The formulation of safety related constraints usually depend on sensory input that gives online information about the dynamic environment the robot interacts with. The controller in this case reactively cope with these constraints as it is impossible to perform a pre-plani cation of safe movements for the robot without accurate data about the present and future dynamics of its environment.
Reactive controllers are usually formulated and solved to convert the operational inputs (user speci ed objectives and constraints) into joint inputs and this, at each time-step. The impact of the retained solution on the solvability of future control problems is rarely considered and therefore, issues of constraints incompatibility may occur. Consequently, this can render the control problem impossible to solve in the future without violating the imposed constraints.
To better understand the problem of constraints incompatibility, the following scenario in which a robot interacts with its dynamic environment is considered a moving telemanipulated robotic arm must keep a safe distance ds from an obstacle O entering its workspace (see Fig. 1.4).
The QP form and the di erent constraints incompatibility cases
This section aims at establishing the general formulation of the constrained, redundant dynamic reactive control scheme. The objective is to compute every time-step the control torque cjk in order to perform a trajectory tracking task in joint space while coping at the same time with the articular position, velocity, torque/acceleration and jerk constraints. We highlight the fact that, desired trajectories for the joints of the robot are not known in advance but discovered at every time-step (tele-operation analogue situation). Incompatibility cases related to the naive way of expressing the articular constraints are exposed and discussed.
Joint position constraint incompatibility with jerk limits 2
As previously explained, to palliate the oscillations induced on the movement of a joint when using (2.20) to cope with an upper or lower position limit, a more complete description of the braking phase must be used. In this case, we consider the braking phase of a joint moving towards its upper position limit qM , the new formulation of the constraint on its position includes both the lower q m and upper q M jerk limits. The implicitly induced braking phase that makes the joint stop at qM is as follows: joint 0 starts jerking negatively with maximum producible jerk q m during n7 iterations to force reduce the amount of acceleration (q 0) it contains. The reached amount of deceleration (q 0) is then progressively \released » and brought to 0 by jerking positively with maximum producible jerk q M during n9 control time-steps (see Fig. 2.13). Maximum producible deceleration qm is not considered in this case. The extended state S of the system during this implicitly induced braking phase can be described as: 8 qjk+1 = qjk + qjk t; 9 S k+1 > q k+1 = q k + q t; >.
Table of contents :
1 Introduction, State-of-the-art and proposed contributions
1.2 Safe mechanical design
1.2.1 Flexible actuators
1.2.2 Lightweight Robotic Systems
1.4 Safety standards
1.5 Discussion and proposed contributions
1.5.1 Considerations when dealing with safety during human-robot interaction
1.5.2 Considerations when dealing with constraints
1.6 Related Publications
2 Constraints incompatibility
2.1.1 The general problem
2.1.2 Dierent types of constraints
2.1.3 Corresponding literature
2.2 The QP form and the dierent constraints incompatibility cases
2.2.1 Task formulation
2.2.2 Controller formulation
2.2.3 Articular constraints: naive formulations
2.3 Test case scenario for simulation
2.4 Articular constraints: new formulations
2.4.1 Joint velocity constraint incompatibility with jerk limits
22.214.171.124 Illustration 1
2.4.2 Joint position constraint incompatibility with deceleration and/or jerk limits
126.96.36.199 Joint position constraint incompatibility with deceleration limits
188.8.131.52.1 Illustration 2
184.108.40.206 Joint position constraint incompatibility with jerk limits 1 .
220.127.116.11.1 Illustration 3
18.104.22.168 Joint position constraint incompatibility with jerk limits 2 .
22.214.171.124.1 Illustration 4
126.96.36.199 Joint position constraint incompatibility with both deceleration and jerk limits 1
188.8.131.52.1 Illustration 5
184.108.40.206 Joint position constraint incompatibility with both jerk and deceleration limits 2
220.127.116.11.1 Illustration 6
2.5 Final bounds on the acceleration control variable
2.5.1 Illustration 7
3 Energy based safe control
3.2 Interaction forces and safety indicators
3.2.1 Kinetic energy
3.2.2 Potential energy
3.2.3 Relation between kinetic and potential energies
3.3 Safety limit values
3.3.1 Safety limit value for the pre-collision safety indicator
18.104.22.168 Pre-collision safety criterion extension
3.3.2 Safety limit value for the physical-contact safety indicator
3.4 Energy constraints implementation
3.5 Task energy prole
3.6 Safe dynamic controller
3.6.1 Task formulation
3.6.2 Constraints formulation
22.214.171.124 Articular constraints
126.96.36.199 Energy related constraints
3.6.3 Controller formulation
3.7.1 Test case scenario
3.7.2 Scenario 1: obstacle intersecting with the trajectory of the robot and no constraints on its energy
3.7.3 Scenario 2: nearby obstacle not intersecting with the trajectory of the robot and constraint on its kinetic energy
3.7.4 Scenario 3: obstacle intersecting with the trajectory of the robot and constraint on its kinetic energy
3.7.5 Scenario 4: obstacle intersecting the trajectory of the robot and constraints on its kinetic and potential energies
3.7.6 Scenario 5: obstacle intersecting with the trajectory of the robot and constraint on its task energy prole
4 Safe human-robot interaction: experimental results
4.2 Experimental setup
4.3 Test case scenario
4.4 Human-robot interaction, scenario 1
4.5 Human-robot interaction, scenario 2
4.6 Human-robot interaction, scenario 3
4.7 Human-robot interaction, scenario 4
5 Conclusions and outlook
5.2.1 Extension of the work on the constraints incompatibility problem
5.2.2 Control of collaborative robots for safe human-robot interaction