Collaborative robots in bioproduction

A conventional industrial robot is fenced because it cannot know where a person is and cannot moderate what it does on contact. A collaborative robot removes the fence by inverting the problem: instead of guaranteeing that contact never happens, it guarantees that contact, when it happens, stays below the threshold at which a human is injured. That threshold is not an engineering convention. ISO/TS 15066 grounds it in experimental pain-onset data for specific body regions, tabulating maximum permissible pressure and force separately for each — the forearm tolerates far more than the face — and distinguishing transient impact from quasi-static clamping, where the body part cannot move away and the tolerable force is much lower.

Why the safe speed changes with the arm’s posture

For a transient collision, what matters is the kinetic energy transferred, and that depends on the arm’s effective mass in the direction of contact rather than on its rated payload. Effective mass is the inertia reflected to the contact point through the current joint configuration: an extended arm swinging about its base presents a large effective mass, while the same arm folded and moving on a wrist axis presents a small one. So a single number for “safe speed” does not exist for a given robot; the limit is pose-dependent, which is why validation is done by measuring force and pressure at the actual contact points of the actual application with an instrumented gauge, not by reading a datasheet.

Detection sets the other half of the behaviour. A robot that senses contact only through motor current has to distinguish a collision from friction, gravity and payload dynamics, all of which appear in the same signal, so the trip threshold must be set high. Joint torque sensing or series-elastic actuation measures the interaction force closer to where it occurs, allowing a lower threshold and a gentler stop. The compliance that makes an arm safe also makes it less stiff, and therefore less precise under load — safety and accuracy are traded, not co-optimised.

The consequence: cobots are not fast

Contact energy scales with the square of velocity, so halving the tolerable energy costs far more than half the speed. A force-limited arm is intrinsically a low-throughput device, and any case built on cycle time against a caged industrial robot loses. What it wins instead is location: it can stand inside an existing bench, an incubator bay or a filling line without the floor area, interlocks and light curtains a fenced cell requires, and it can work in a space a human also uses.

What bioproduction adds

The reason to automate an aseptic step is rarely labour cost. It is that the operator is the dominant source of viable contamination, the same argument that produces aseptic isolators — and a robot inside a barrier removes interventions rather than staff. That imposes constraints unrelated to safety: moving joints and belts shed particles, lubricants outgas, and every external surface must survive repeated vaporised hydrogen peroxide exposure without crazing, which favours sealed, smooth-skinned arms over cable-exposed ones.

The harder residual problem is not the arm but what it holds. Biological work involves deformable, wet, tolerance-variable objects — tubing, bags, caps, plates from different mould lots. Most failures in laboratory automation trace to consumable variation rather than to robot error, which is why force-feedback grippers and vision matter more here than repeatability figures.