Illustrative diagram — the operator (exception handler) and the AMR fleet feed into a single system KPI dashboard, not isolated metrics.
❌ The classic mistake: individual UPH with AMRs
The most common legacy metric in warehouses is Individual Units Per Hour (UPH): counts how many units each operator processes per hour. In manual operations, it made sense. With AMRs (autonomous mobile robots) in operation, it becomes actively harmful — penalizing exactly those who do the most critical work.
🎯 The perverse incentive
The operator who stops to resolve a robot exception loses UPH — but without it, the entire fleet stops. The metric rewards those who ignore the problem and perform simple tasks, and penalizes those who keep the system running.
- •Measures the part, ignores the whole
- •Penalizes the exception handler — the highest-value role
- •Creates competition where there should be collaboration between people and the fleet
✓ What the metric SHOULD reward
- ✓Keep the fleet running continuously
- ✓Resolve exceptions quickly and well
- ✓Total output of the human + robot system
✗ What individual UPH does
- ✗Counts units of ONE isolated operator
- ✗Punishes those who stop to save the fleet
- ✗Rewards simple tasks over critical work
Key concepts
📈 System Throughput: Overall System Output
O System Throughput measures the total output of the human-robot system as a integrated unit — not the sum of the individual parts. The value of modern operations isn’t in how much each person carries; it’s in synergy between the person operating it and the fleet they maintain.
📊 The data that changes the conversation
- •An operator who maintains 10 AMRs running generates more throughput than 10 operators doing manual picking.
- •The benefit comes from human-machine synergy, not the sum of individual effort.
💡 Practical tip
Before evaluating a person, ask: “What was the system's output with them present versus absent?” If the fleet's performance drops without the operator, their throughput is enormous—even if their individual UPH is low.
Key concepts
🔧 Exception Rate: Interventions per Hour
O Exception Rate counts the number of human interventions needed per hour of robotic operation. It's the metric that measures integration quality between human and robot—and it’s the most revealing measure of the system’s maturity.
🔍 What a High Exception Rate Is Telling You
A high exception rate isn’t the operator’s fault—it’s a symptom. It points to upstream problems:
- •Poor setup: poorly configured layout, signage, or docking
- •Insufficient training: the operator doesn't prevent problems, only puts out fires
- •Flow design: routes that cause collisions or recurring blockages
✓ Use the exception rate to
- ✓Find the root cause of recurring exceptions
- ✓Measure integration maturity over time
- ✓Prioritize where to invest in setup and training
✗ Do NOT use the exception rate to
- ✗Blame the operator who handles the exceptions
- ✗Ignore the root cause and just demand the numbers
- ✗Compare areas with different layouts without context
Key concepts
⏱️ Fleet Uptime and MTTR: operator responsibility
Two metrics that finally value the human role: Fleet Uptime (% of time the fleet operates normally) and MTTR — Mean Time to Resolve (how long the operator takes to resolve each exception). Uptime is the operator’s direct responsibility; MTTR reveals competence—and it can be trained.
Fleet Uptime %
Percentage of time the fleet operates normally
Direct responsibility of the operator: the faster they prevent and resolve issues, the more productive the fleet is. This is the fleet operator’s central KPI.
MTTR for exceptions
Average time to resolve each exception
Demonstrates problem-solving skills — and, unlike “talent,” they can be trained. A trained operator lowers MTTR and increases uptime at the same time.
💡 Practical tip
Uptime and MTTR are linked: whoever reduces MTTR almost always increases uptime. Train people to resolve specific exceptions (docking, blockage, code reading), and both metrics improve together—that’s the concrete case for reskilling delivering measurable returns.
Key concepts
🎯 Zero errors in handoffs: critical transitions
Handoffs are the handoff points between humans and robots — and the moments of greatest risk in hybrid operations. The goal is zero failures (zero-defect target) at these specific points, because that's where accidents and losses are concentrated.
⚠️ Why the handoff is the weak point
Accidents and losses occur predominantly during handoffs — when responsibility shifts from a human to a robot or vice versa. Measuring and training for these specific transition points drastically reduces risk.
Identify each critical handoff
Map every point where control/load passes between a human and a robot. Measure each handoff point.
Define zero-defect target
Set a zero-failure target for each handoff and count deviations. Unlike an average-based metric, every failure counts here.
Transition-specific training
Train the exact handoff gesture—not the job in general. Focusing training on the point of risk reduces failures where they’re most costly.
Key concepts
🧠 92.5% Framework: Intent Recognition
The next generation of HRC (Human-Robot Collaboration) metrics starts with robots that understand the human intent. A research framework (arxiv, 2025) reaches 92.5% accuracy in recognizing human intentions in collaborative environments — and that changes what can be measured.
📊 Research data
- •92.5% accuracy in human intent recognition (arxiv 2025).
- •Robots that anticipate intent reduce exceptions before they happen.
- •More natural and safer collaboration—fewer failed handoffs.
🚀 Why this is the future of HRC metrics
When the robot recognizes intent, the metrics from the previous topics improve in a cascade: fewer exceptions (topic 3), more uptime (topic 4), and safer handoffs (topic 5). Intent recognition is the upstream lever for almost everything.
- •Reduces exceptions at the source, not just during resolution
- •Increases safety at transition points
- •Makes human-robot collaboration more natural
💡 Practical tip
Use the framework as a horizon, not a prerequisite. Start by measuring throughput, exception rate, uptime/MTTR, and handoffs today. Intent recognition (92.5%) is the next step that makes these metrics structurally better.
Key concepts
📌 Module Summary
Next Module:
3.3 - Business Case: Making the Case for Reskilling in Robotics