Why Science? | National Robotics Programme

Official NRP pages connect capability development, embodied AI, industry translation, standards, events and ecosystem growth. The programme convenes actors and supports work at different maturity levels; individual projects retain their own evidence and responsibilities.

Embodied AI must work through sensors, actuators and physical contact. The real world adds friction, occlusion, latency, wear and people, turning an elegant algorithm into a systems-engineering problem. The practical capability is institutional literacy. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “NRP is Singapore’s national robotics platform” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: National support does not certify every robot or vendor.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Next chapter

Embodied AI connects perception, reasoning and learning with a body that moves and affects the environment. Sensors, compute, power, actuators and mechanical design constrain one another. Intelligence should therefore be assessed through the complete task loop.

A robot should be evaluated on the task, environment and failure conditions it will face. Performance in simulation is useful evidence, while safe real-world trials provide a different and necessary layer. The practical capability is systems thinking. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “Embodied AI acts through a physical system” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: A language or vision model alone is not an operational robot.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

A useful specification names the user, setting, objects, timing, constraints, hazards and acceptable failure rate. Observing the present workflow often reveals requirements that a broad goal hides. Alternatives should be compared before deciding a robot is appropriate.

Robotics standards help teams describe interfaces, hazards, tests and responsibilities consistently. Interoperability and safety are engineering achievements that support adoption rather than paperwork added afterwards. The practical capability is requirements engineering. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “Task definition comes before technology choice” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: A technically possible automation may still be a poor workflow choice.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

Cameras, depth sensors, force sensors, encoders and other instruments measure different variables with noise and blind spots. Lighting, occlusion, surface properties and calibration affect observations. Sensor fusion combines evidence without creating certainty from bad inputs.

Translation succeeds when users help define the problem and can operate, maintain and trust the resulting system. Technical capability, workflow design and human factors belong in the same evidence trail. The practical capability is sensor literacy. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “Sensors provide partial views of the world” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: More sensors do not automatically produce an accurate state estimate.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

Models may detect objects, estimate pose, track motion or recognise events. Training data, labels and environment coverage shape performance. Confidence scores need calibration, and failure sets should include rare or ambiguous scenes.

Embodied AI must work through sensors, actuators and physical contact. The real world adds friction, occlusion, latency, wear and people, turning an elegant algorithm into a systems-engineering problem. The practical capability is perception evaluation. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “Perception turns signals into task information” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: Benchmark accuracy does not establish reliability in a new workplace.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

A robot combines current measurements with prior motion and models to estimate where it and relevant objects are. Drift, latency and missing observations accumulate. Uncertainty helps the planner decide when to slow, re-observe or stop.

A robot should be evaluated on the task, environment and failure conditions it will face. Performance in simulation is useful evidence, while safe real-world trials provide a different and necessary layer. The practical capability is probabilistic reasoning. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “State estimation keeps a changing world coherent” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: An estimated pose is not the exact physical state.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

A planner searches for paths or action sequences that meet goals while respecting geometry, time and rules. Long-horizon tasks amplify small mistakes. Replanning helps with change, but only if the system can detect that its assumptions failed.

Robotics standards help teams describe interfaces, hazards, tests and responsibilities consistently. Interoperability and safety are engineering achievements that support adoption rather than paperwork added afterwards. The practical capability is planning literacy. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “Planning selects actions under constraints” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: A collision-free plan in a model may be unsafe in an altered scene.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

Controllers compare desired and measured motion and update commands. Dynamics, friction, payload, actuator limits and delays affect stability and precision. Safe limits and fallback modes matter as much as nominal tracking performance.

Translation succeeds when users help define the problem and can operate, maintain and trust the resulting system. Technical capability, workflow design and human factors belong in the same evidence trail. The practical capability is control-systems reasoning. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “Control closes the loop” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: Good trajectory tracking does not prove safe task behaviour.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

Grasping, inserting, wiping or assembling require perception and force interaction. Objects may deform, slip or vary. Compliance and tactile feedback can help, while fixtures and process redesign may simplify the problem more reliably than a clever policy.

Embodied AI must work through sensors, actuators and physical contact. The real world adds friction, occlusion, latency, wear and people, turning an elegant algorithm into a systems-engineering problem. The practical capability is manipulation science. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “Manipulation combines contact, geometry and uncertainty” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: Success on one object set does not establish general manipulation capability.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

Mobile robots map space, localise and avoid obstacles, but human environments also contain conventions, visibility limits and changing priorities. Comfortable distance and predictable motion affect trust. Site rules should define right of way and safe stops.

A robot should be evaluated on the task, environment and failure conditions it will face. Performance in simulation is useful evidence, while safe real-world trials provide a different and necessary layer. The practical capability is human-aware navigation. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “Navigation is social as well as geometric” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: Reaching a destination does not prove socially acceptable navigation.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

Simulation supports repeatable scenarios, synthetic data and safer early testing. The physics, sensors and behaviour models remain approximations. Domain randomisation and calibration can reduce gaps while real trials still test what the simulator omitted.

Robotics standards help teams describe interfaces, hazards, tests and responsibilities consistently. Interoperability and safety are engineering achievements that support adoption rather than paperwork added afterwards. The practical capability is sim-to-real reasoning. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “Simulation accelerates learning with limits” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: Simulation success is not equivalent to physical deployment evidence.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

Task completion, cycle time, accuracy, intervention, energy, recovery and safety may all matter. A useful test suite includes normal cases, edge cases and changed conditions. Baselines reveal whether complexity creates a practical improvement.

Translation succeeds when users help define the problem and can operate, maintain and trust the resulting system. Technical capability, workflow design and human factors belong in the same evidence trail. The practical capability is evaluation design. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “Evaluation should expose failures” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: A curated demonstration cannot represent the operating distribution.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

Risk assessment considers hazards, people, energy, tools, unexpected motion and foreseeable misuse. Protective measures can include inherently safe design, guarding, sensing, speed limits, procedures and emergency response. Residual risk should be communicated.

Embodied AI must work through sensors, actuators and physical contact. The real world adds friction, occlusion, latency, wear and people, turning an elegant algorithm into a systems-engineering problem. The practical capability is robot-safety literacy. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “Safety is designed into the system” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: A robot labelled collaborative is not safe in every application.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

NRP says it leads robotics-related standards work in Singapore and participates in international committees. Standards can define vocabulary, interfaces, test methods and safety practices. The applicable edition and scope must be checked for each system.

A robot should be evaluated on the task, environment and failure conditions it will face. Performance in simulation is useful evidence, while safe real-world trials provide a different and necessary layer. The practical capability is standards literacy. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “Standards create shared expectations” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: A standards list does not establish that a specific deployment conforms.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

Robots interact with lifts, doors, fleet managers, factory equipment and information systems. Common interfaces can reduce custom integration and vendor lock-in. Cybersecurity, authentication and failure behaviour belong in interface design.

Robotics standards help teams describe interfaces, hazards, tests and responsibilities consistently. Interoperability and safety are engineering achievements that support adoption rather than paperwork added afterwards. The practical capability is interoperability reasoning. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “Interoperability expands useful ecosystems” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: Technical connectivity does not by itself authorise access or data sharing.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

NRP’s RoboCluster page names six sectors and describes collaboration among public R&D institutions, end users, robotics companies and agencies. Sector knowledge shapes workflows, constraints and adoption. Translation should preserve the original evidence when a prototype changes hands.

Translation succeeds when users help define the problem and can operate, maintain and trust the resulting system. Technical capability, workflow design and human factors belong in the same evidence trail. The practical capability is translation literacy. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “RoboClusters connect research with sector needs” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: Participation in a cluster does not guarantee a commercial outcome.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter · Next chapter

Archive task version, environment, hardware, firmware, datasets, models, calibration, simulator, test scenarios, interventions, failures, safety review, user feedback and maintenance state. Students can build safe tabletop systems while qualified teams control workplace deployments.

Embodied AI must work through sensors, actuators and physical contact. The real world adds friction, occlusion, latency, wear and people, turning an elegant algorithm into a systems-engineering problem. The practical capability is robotics career preparation. Keep the question, method, controls, observation, uncertainty, review and next decision in one evidence receipt. That record makes progress visible without enlarging the claim beyond the stage reached. NRP and its partners control current programmes, funding, events and released results; deployers, safety owners, standards bodies and regulators control real-world operation.

Put this into practice by turning the idea into a small, reviewable action: Build a dated evidence card for “Keep a robotics evidence receipt” with the question, method, observation, limitation and next test. Then ask what the resulting note or artefact actually demonstrates. A useful record names context, source or instruction, action, observation, interpretation and uncertainty. It also preserves this limit: Only current responsible owners can approve operation around people and equipment.

An evidence check keeps confidence proportional to what the work can show. Separate the original observation or official statement from the learner’s explanation. Record feedback and any revision. If information is missing, write “not shown” or “not yet known” rather than guessing. That honest gap can become the next question.

Contents · Previous chapter