This lecture explains how advances in Artificial Intelligence (AI) influence the development and deployment of modern robots, which tasks robots may take on in the future, where the central technical and societal challenges lie, and how robotics can change our everyday lives. Statements are based on the publications listed in the sources; uncertainties and data gaps are indicated where applicable.
Duration: approx. 20 minutes
Robotics combines mechanics, sensing, control and increasingly AI-supported software. A helpful analogy is that of a human: motors and joints correspond to muscles and bones, sensors correspond to eyes, ears and skin, control and planning correspond to the brain. AI adds new capabilities to the "brain": more flexible perception, adaptive planning and learning from experience. The Springer Handbook of Robotics systematically presents these core components and their interconnections and remains a central technical reference framework (Siciliano & Khatib, eds.).
The impact of AI is visible in concrete examples: In industrial logistics centers, automated vehicles and robotic arms today move much more flexibly than early industrial robots; they use cameras and learning algorithms to recognize objects and adjust grasps. Companies such as Amazon have transformed logistics through automated conveyor and gripping systems; information on this application area and technical approaches can be found on the Amazon Robotics pages.
In the medical field, robotic surgical systems such as the da‑Vinci system assist surgeons by enabling finer movements and stabilizing incisions. These systems combine mechanical precision with imaging support; manufacturers document these applications in their product descriptions (Intuitive Surgical).
A central quality feature of modern robots is the ability to learn from examples. In imitation learning, a robot is shown how to perform a task by observation or teleoperation; it can then generalize from this. This method is systematically summarized in survey articles (Argall et al., 2009) and is particularly suitable when explicit programming would be very laborious.
In summary: AI complements classical control and planning methods with data-driven perception and adaptation. This changes application domains, makes systems more flexible, but also brings new uncertainties (for example regarding robustness outside the training data).
Duration: approx. 20 minutes
To methodically understand the changes brought by AI, we introduce key terms and classify them according to tasks. Perception refers to the processing of sensor data — e.g., camera images, LiDAR or tactile sensing. In research, learning-based image processing models complement classical signal processing, especially when environments are variable. For transferring perception and control models learned in simulation to the physical world, the method of "domain randomization" is known: by randomly varying the simulated scene during training, a model can become more robust to real-world deviations (Tobin et al., 2017).
Planning refers to finding a sequence of movements or states that achieves a goal. Traditional planners work with models of kinematics and dynamics; AI methods add learning-based components here, e.g., value estimates in reinforcement learning control or policy networks that map directly from images to actions. Imitation learning (Learning from Demonstration) is a pragmatic approach to obtain policies from examples; a comprehensive overview can be found in Argall et al. (2009).
Control is the fine-grained execution of planned trajectories into motor commands while accounting for disturbances. Hybrid systems combine model-based feedback with data-driven adjustments to maintain safety and stability (see Springer Handbook of Robotics).
Further important terms: Sim-to-Real refers to the transfer problem of models trained in simulation to the real world (see Tobin et al., 2017). Manipulation denotes grasping and handling of objects; for complex, deformable or variable objects the problem of robust grasp planning remains open and is an active research area (Siciliano & Khatib, eds.). Human-Robot Interaction (HRI) encompasses methods by which robots recognize human intentions, communicate understandably and execute tasks jointly. Safety-related approaches and ethical guidelines are discussed in political and professional documents; the EU Guidelines for Trustworthy AI summarize principles and requirements for transparency, robustness and accountability (European Commission, HLEG, 2019).
It is important to acknowledge possible limitations of learning-based components: training data are never fully representative of all real situations. Therefore, robustness testing, formal verification for safety-critical subfunctions and monitoring mechanisms are essential complements. The Springer Handbook of Robotics and the EU guidelines provide methodological advice and principles for this.
In conclusion: The combination of perception, planning and control with data-driven methods allows robots to perform tasks far more flexibly than before. At the same time, some challenges shift from model specification to ensuring robustness and reliability in unknown environments.
Duration: approx. 10 minutes
Concrete applications that already benefit from AI today include industrial logistics (sorting, transporting containers), assistive care (support with mobility, accompaniment), operative assistance in surgery (precision manipulation with visual support) as well as inspection tasks in the energy and infrastructure sectors. Manufacturers and providers describe typical use scenarios: Amazon Robotics for logistics solutions and Intuitive Surgical for robot-assisted surgery provide practical insights into concrete systems.
At the same time, clear limits remain: Robots are still constrained in very unstructured, variable everyday scenarios. Tasks that require basic everyday manipulations with unfamiliar objects and complex social interaction remain challenging, because training data rarely cover all edge cases and physical interactions produce many unexpected contact states. Research on transferring simulation experience to the real world (Sim-to-Real) is an active attempt to reduce this gap (Tobin et al., 2017), but complete solutions do not yet exist.
Thought exercise 1: Imagine a service robot that should pick up plates in a private apartment, sort dishware into a dishwasher, and find items at varying locations. Which components would you prioritize: robust object recognition, tactile sensing for variable grasp points, situational planning, or interaction with the residents? Discuss which risks can arise if only one component is strong (e.g., good recognition but poor grasp control).
Thought exercise 2: A warehouse wants to decide whether investing in learning-capable gripping arms is worthwhile. Which criteria would be economically and technically relevant (e.g., variance of objects to be grasped, frequency of new object classes, requirements for downtime)? Which uncertainties would they need to quantify?
These tasks show: technical feasibility is only part of the decision; robustness, operational conditions as well as ethical and legal questions play a major role. The EU guidelines provide examples of governance and safety requirements that should be considered during deployment.
In closing, it should be emphasized again: the integration of AI into robotics expands the range of possible tasks and increases flexibility, but poses new requirements for researchers and operators regarding robustness, testability and responsible design. More precise forecasts on adoption and macroeconomic effects are the subject of ongoing studies and vary by sector; there are data gaps and uncertainties in this area.