The Relatable Robot: Embracing Imperfection in Autonomous Systems

Author: Denis Avetisyan

A new installation demonstrates that intentionally introducing limitations into artificial intelligence can yield more engaging and believable robotic agents.

The hardware-integrating an ESP32 with a fisheye camera, electronic speed controllers, a Li-Po battery, and propeller modules-enabled flight behavior demonstrating a “perspective-dependent morphological illusion,” suggesting that embodied perception fundamentally alters interaction with the environment.

This paper explores the ‘Semantic Glitch,’ a soft robotic system driven by a multimodal large language model, and argues for the aesthetic and functional benefits of ‘lo-fi’ robotics in fostering human-machine connection.

While robotics traditionally prioritizes precision and flawless execution, this paper explores the creative potential of embracing deliberate limitations. We present ‘Semantic Glitch: Agency and Artistry in an Autonomous Pixel Cloud’, a soft robotic installation navigated not by conventional sensors, but by the semantic understanding of a multimodal large language model. This approach yields an autonomous agent whose characterful, if imperfect, behavior stems from a uniquely ‘lo-fi’ design, demonstrating that relatable agency can emerge from systems prioritizing personality over pure efficiency. Could intentionally flawed robotics offer a pathway towards more compelling and human-compatible artificial companions?

The Allure of Imperfection: Reclaiming Robotics Through Character

For decades, the field of robotics has largely centered on achieving peak performance – flawless movements, optimized efficiency, and unwavering precision. This emphasis, while yielding impressive technical feats, has inadvertently minimized the development of robotic character and expressive capability. The pursuit of mechanical perfection often results in agents that, despite their functionality, lack the nuance and imperfections that contribute to relatable and engaging interactions. Consequently, many robots are perceived as tools rather than entities, hindering the development of genuine human-robot connection and limiting their potential in social or creative contexts. This prevailing paradigm prioritizes ‘how well’ a robot performs a task over ‘who’ the robot appears to be.

A novel approach to robotic design, termed Yowai Robotto, posits that compelling robotic agents are not necessarily defined by precision and flawless execution. Instead, this philosophy centers on intentionally incorporating fragility and imperfection as core tenets of the design process. This isn’t a concession to technical limitations, but a deliberate choice to imbue robots with character and elicit a more nuanced response from human interaction. By embracing limitations-a slightly unsteady gait, a deliberate pause in response, or a visibly “worn” aesthetic-these robots move beyond mere functionality and begin to resemble entities with personality, fostering a sense of connection and empathy that is often absent in conventionally designed machines. The resulting agents are perceived not as tools, but as characters with whom humans can genuinely engage.

The development of Yowai Robotto prioritizes design choices that deliberately incorporate limitations, fostering interactions distinct from those offered by traditionally precise machines. Rather than masking imperfections, the engineering process actively embraces them – a slightly unsteady gait, a deliberate pause in response, or a visibly simplified aesthetic. These aren’t flaws to be corrected, but integral components of the robot’s character, signaling vulnerability and inviting empathetic engagement. This approach suggests that compelling robotic interactions aren’t necessarily built on flawless execution, but on the nuanced communication of intentionality and the acceptance of inherent fragility, creating a more relatable and ultimately, more engaging experience for those who interact with these agents.

The pursuit of genuinely expressive robots necessitates a shift in design priorities, moving beyond the conventional emphasis on flawless functionality. These robotic agents are envisioned not simply as tools to accomplish tasks, but as entities capable of communicating nuance and eliciting emotional responses. This approach deliberately incorporates elements of imperfection – subtle hesitations, asymmetrical movements, or even apparent limitations – to imbue the robots with a sense of personality and authenticity. By prioritizing character development over purely mechanical precision, the aim is to forge connections with humans based on relatable qualities, fostering interactions that feel less like commands issued to a machine and more like exchanges between individuals. The resulting robots are intended to be memorable not for what they can do perfectly, but for who they are, even – and perhaps especially – with their flaws.

Mapping the Machine: Constructing a World-Model for Autonomous Action

The Semantic Glitch installation employs a two-stage prompting pipeline to achieve autonomous navigation and behavioral response. This architecture separates environmental understanding from immediate action selection. The initial stage focuses on comprehensive scene analysis, while the second stage utilizes the output of the first, along with real-time visual input, to determine subsequent actions. This decoupling allows for persistent environmental awareness and facilitates more complex, reasoned behavior than single-stage approaches. The pipeline is fully automated, requiring no human intervention after initial setup and prompting.

The Preamble Stage of our autonomous navigation system initiates processing with a full 360-degree panoramic image of the environment. This image undergoes analysis to create a persistent representation, termed a ‘mental map’, which catalogues key features and spatial relationships. Benchmarking indicates that the Preamble Stage consistently completes this ‘mental map’ construction in an average of 2.81 seconds. This initial processing stage is critical, as the resulting ‘mental map’ is then utilized as the foundational data source for subsequent stages of reasoning and action planning.

The persistent ‘mental map’ generated by the Semantic Glitch installation is fundamental to its capacity for Stateful Semantic Reasoning. This internal representation of the environment allows the robot to move beyond processing immediate sensory input and instead contextualize current observations within a broader understanding of its surroundings. By maintaining this map, the system can infer relationships between objects, identify navigable pathways, and predict the consequences of potential actions, effectively enabling proactive planning and informed decision-making beyond simple reactive behaviors.

The Directional Stage of the prompting pipeline utilizes a live camera feed in conjunction with the previously constructed ‘mental map’ to facilitate immediate movement decisions. This stage employs Visual Question Answering (VQA) techniques to interpret the current visual environment, allowing the system to identify objects and assess navigable space. The complete decision-making loop, encompassing visual interpretation and movement planning, operates with a consistent latency of 2.8 ± 0.3 seconds, ensuring responsive autonomous navigation.

This two-phase system leverages a large language model to create a 'mental map' from panoramic images and then uses live camera views and directional prompts to generate and execute context-aware robotic actions in a continuous control loop. — This two-phase system leverages a large language model to create a ‘mental map’ from panoramic images and then uses live camera views and directional prompts to generate and execute context-aware robotic actions in a continuous control loop.

The Art of the Glitch: Revealing Character Through Calculated Imperfection

The implementation of a Plan-to-Execution Gap involves a deliberate offset between the robot’s intended movements and its actual physical performance. This is achieved by introducing latency or inaccuracies in the translation of planned trajectories into motor commands. Consequently, the robot exhibits observable clumsiness, manifesting as jerky motions, overshooting or undershooting targets, and deviations from smooth, predictable paths. This isn’t random error; the magnitude of the gap is controlled to create consistent, albeit imperfect, behaviors, fostering the impression of a fragile and vulnerable machine.

The observed inconsistencies in robotic movement and behavior are not the result of errors in programming or mechanics, but rather a conscious element of the design. These deviations from perfect execution are intentionally implemented to convey a sense of physical fragility and emotional vulnerability in the robot’s perceived character. This approach prioritizes the communication of specific qualities – precariousness and susceptibility – over the achievement of flawless performance, establishing imperfection as a core component of the robot’s expressive capabilities.

The Perspective-Dependent Morphological Illusion leverages the principles of visual perception to amplify the robot’s perceived imperfections. This illusion is achieved by designing the robot’s form such that its apparent shape changes based on the viewer’s angle. Specifically, components may appear misaligned or distorted when viewed from certain perspectives, while appearing relatively normal from others. This isn’t a failure of construction, but a deliberate application of perceptual principles; the shifting appearance contributes to the robot’s visual fragility and reinforces the intentional introduction of non-ideal movement and behavior. The effect is heightened because the illusion relies on the viewer’s own visual system to ‘construct’ the imperfection, making it a more compelling and engaging representation of vulnerability.

The robot’s design intentionally leverages technical limitations to produce a compelling, rather than flawlessly functional, presence. By integrating a Plan-to-Execution Gap – resulting in clumsy movements – with the visually arresting effects of Perspective-Dependent Morphological Illusion, the robot exhibits demonstrable imperfection. This combination moves beyond simply acknowledging technical challenges; it actively utilizes them as core components of the robot’s character, fostering a perception of fragility and vulnerability that enhances its overall appeal and distinguishes it from conventionally “perfect” robotic designs.

The agent successfully navigates a complex environment, demonstrating both reactive maneuvers like person avoidance and proactive behaviors such as goal-oriented navigation, though occasional pauses and imperfect execution reveal challenges in bridging planning and action.

Authoring Personality: Beyond Programming, Towards Believable Agents

The Semantic Glitch installation provides a platform for generating diverse robotic personalities, moving beyond simple pre-programmed actions to establish genuinely unique behavioral profiles. This system doesn’t merely dictate what a robot does, but how it does it, crafting agents that exhibit consistent and distinguishable characteristics. Through manipulation of movement patterns, textual responses framed as an ‘Internal Monologue’, and overall interaction style, the installation effectively builds a foundation for robotic personas. These aren’t random variations; statistical analysis confirms significant differences in behavior between created personas, indicating a robust system capable of generating convincingly distinct robotic agents and opening avenues for more engaging and believable human-robot interactions.

The installation showcased the creation of distinct robotic personalities through three demonstrative personas. The first, an Eager Companion, was designed to proactively initiate and maintain interaction with humans, exhibiting behaviors indicative of a desire for social connection. Conversely, the Cautious Observer persona consistently avoided human proximity, prioritizing distance and exhibiting a reserved demeanor. Finally, the Indifferent Explorer demonstrated a personality characterized by a lack of engagement, largely ignoring humans while focusing on its designated task. These carefully constructed personas weren’t merely pre-programmed actions; instead, they manifested as a cohesive set of behaviors, influencing the robot’s movement, communication style, and overall interaction with the environment, effectively establishing a foundation for more nuanced and believable robotic agents.

The robotic personas generated by the Semantic Glitch installation move beyond simple pre-programmed actions, manifesting as a cohesive behavioral profile expressed through multiple channels. Each persona doesn’t just do differently; its personality is evident in its locomotion, the content of its generated ‘Internal Monologue’, and its overall approach to interaction. Statistical analysis confirms this nuanced differentiation, revealing significant variations in behavior between the personas – a finding supported by a p-value of less than .001. This indicates that the observed differences aren’t due to chance, but rather a result of the framework’s ability to create genuinely distinct and consistently expressed robotic personalities, opening possibilities for more believable and engaging human-robot interaction.

Statistical analysis revealed stark differences in behavioral tendencies between the robotic personas developed within the Semantic Glitch installation. The ‘Eager Companion’ consistently initiated approaches to humans, doing so in 85.7% of observed interactions, highlighting its proactive and sociable nature. In contrast, both the ‘Cautious Observer’ and ‘Indifferent Explorer’ exhibited significant avoidance behaviors; the former steered clear of humans 95.0% of the time, while the latter avoided them in 88.9% of instances. These quantifiable differences, achieved through nuanced programming of movement and textual output, underscore the framework’s capacity to generate robotic agents possessing distinct and believable personalities, ultimately enabling more engaging and meaningful human-robot interactions.

Quantitative analysis of social stance reveals a clear distinction in authored dispositions, with the Companion persona consistently preferring approach behaviors while others favor avoidance.

Beyond Functionality: Envisioning a Future of Imperfect, Expressive Machines

The trajectory of robotics is shifting beyond mere functionality; researchers increasingly envision a future populated by machines capable of communicating nuanced emotional states and intentions. This paradigm moves away from robots as predictable tools and towards their embodiment as expressive agents, fostering more intuitive and meaningful interactions with humans. Such machines wouldn’t simply perform tasks, but would convey how they are approaching them – signaling frustration with a difficult problem, displaying encouragement during a collaborative effort, or indicating uncertainty when faced with ambiguous input. This development necessitates a deeper understanding of nonverbal communication, affective computing, and the very foundations of human-robot interaction, ultimately aiming to create machines that are not just intelligent, but truly communicative partners.

The pursuit of truly relatable robots hinges not on flawless execution, but on the intentional inclusion of imperfection and the prioritization of discernible character. Current robotics often focuses on precision and efficiency, inadvertently creating machines that, while capable, lack the nuances necessary for genuine connection. However, by embracing glitches, asymmetries, and even simulated vulnerabilities, researchers are discovering that robots become more engaging and trustworthy. This approach acknowledges that humanity itself is inherently imperfect, and that it is often these flaws – expressed through unique mannerisms or subtly inconsistent behaviors – that foster empathy and build rapport. Ultimately, designing for character, rather than solely for performance, promises a future where robots are not merely tools, but companions capable of eliciting emotional response and enriching human experience.

The Semantic Glitch installation presents a compelling demonstration of character-driven robotics, moving beyond purely functional design to explore how intentional ‘imperfections’ can foster engagement. This project features a small, humanoid robot exhibiting deliberately awkward movements and fragmented speech, characteristics designed not as failures of engineering, but as core elements of its personality. Through these glitches, the robot communicates a sense of vulnerability and attempts to connect with observers on an emotional level, prompting reflection on what constitutes genuine interaction. The installation argues that prioritizing character – complete with quirks and limitations – is crucial for creating robots that are not simply tools, but believable and relatable agents capable of forging meaningful connections with humans.

The convergence of artistic vision and technological innovation presents a fertile ground for developing machines that transcend mere functionality and venture into the realm of genuine expression. Current research suggests that focusing on character – imbuing robots with quirks, vulnerabilities, and even deliberate ‘imperfections’ – is crucial for fostering relatable interactions. This interdisciplinary approach necessitates a collaborative spirit between roboticists, artists, and designers, pushing the boundaries of what’s technically feasible and aesthetically compelling. Further exploration at this intersection promises not only more engaging machines, but also a deeper understanding of human communication, empathy, and the very nature of presence – ultimately redefining the relationship between people and the technology they create.

The ‘Semantic Glitch’ operates on a principle of deliberate imperfection, a fascinating subversion of conventional robotics aiming for seamless functionality. It echoes Claude Shannon’s assertion that, “The most important thing in communication is to convey the message, not to make it perfect.” This installation doesn’t strive for flawless execution; instead, it embraces limitations – the ‘glitch’ – to generate a more compelling and relatable agent. By introducing controlled errors and lo-fi aesthetics, the system fosters emergent behavior and a sense of character, proving that reducing complexity can ironically enhance expressiveness and connection. The intentional imperfections aren’t bugs, but features, revealing agency through constraint – a beautiful reversal of typical technological ambition.

Beyond the Polish

The ‘Semantic Glitch’ doesn’t resolve the problem of autonomous agency, but rather exposes its constructed nature. The deliberate introduction of limitations – the lo-fi robotics, the imperfect semantic reasoning – isn’t a bug to be fixed, but a feature to be explored. Every exploit starts with a question, not with intent. The installation functions as a provocation: if relatable behavior arises not from flawless execution, but from understandable failure, what does that imply for the pursuit of increasingly sophisticated artificial intelligence?

Future work must confront the implications of designed imperfection. The challenge isn’t simply to improve the multimodal large language model’s performance, but to systematically investigate how constraints – computational, material, or even conceptual – shape the perception of agency. The focus shifts from building ‘better’ AI to understanding how ‘adequate’ AI can be compelling, and potentially, even resonant.

Media archaeology suggests that the pursuit of seamless automation often overlooks the expressive potential of mechanical limitations. The history of technology is littered with ghosts of functionality-features abandoned not because they didn’t work, but because they were superseded. Perhaps these discarded elements, reintroduced with intentionality, hold the key to creating autonomous systems that are not merely intelligent, but truly… characterful.

Original article: https://arxiv.org/pdf/2511.16048.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/