Full text
| |
| Author(s): |
Letícia Mara Berto
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Instituto de Computação |
| Defense date: | 2025-03-17 |
| Examining board members: |
Esther Luna Colombini;
Diana Francisca Adamatti;
João Eduardo Kögler Junior;
Carlos Henrique Costa Ribeiro;
Sandra Eliza Fontes de Avila
|
| Advisor: | Esther Luna Colombini; Ricardo Ribeiro Gudwin |
| Abstract | |
The integration of robots into daily activities is growing, necessitating their adaptation to increasingly complex environments. This adaptation requires robots to develop learning capabilities through interaction with their surroundings and other agents, enabling them to perform tasks beyond pre-programmed instructions. Central to this is cognition, the ability to transform information from diverse sources into actionable knowledge through interconnected cognitive functions. Our research is grounded in developmental and cognitive robotics, inspired by human development processes, and contributes to the understanding of how robots integrate motivation, decision-making, and adaptability to function autonomously and engage meaningfully with humans and other robots in dynamic environments. Our initial experiments focused on the robot's interaction with its environment, while later studies incorporated social interaction. A core challenge in robotics lies in replicating decision-making. So, we adapted a human decision-making experiment involving uncertainty, rewards, and punishments to a robotic context. Using reinforcement learning, we trained simulated robots to maintain homeostasis. Results indicated that robots adopting long-term strategies not only survive longer but also better satisfy diverse needs. Beyond addressing basic needs, humans develop preferences shaped by perceived pleasure, adding complexity to decision-making. To model this, we extended Hull’s motivational theory by incorporating hedonic dimensions, enabling robots to balance needs and preferences based on context. We trained three agents with distinct metabolic rates in varied environments. Our findings revealed that agents adapted most effectively when environmental conditions aligned with their metabolic requirements. Moreover, introducing pleasure into the motivational framework influenced behavior, particularly in agents with moderate metabolic rates. Under survival pressures, agents prioritized immediate needs over pleasure, highlighting the dynamic interplay between motivation, pleasure, and environmental context in decision-making. To explore human-robot interaction, we developed a cognitive architecture inspired by young children’s needs, incorporating three drives: learning, interaction, and recharging. By adjusting the importance of these drives, we created two robot profiles: Playful and Social. Experiments demonstrated that changes in drive importance resulted in distinct robot behaviors and human perceptions. Participants adapted to the robots’ behaviors, attributing traits that aligned with their profiles, even without prior knowledge of them. Finally, we examined interactions between two robotic agents with varying needs, employing theory of mind to estimate and respond to each other's requirements. Agents balanced helping themselves or their partners based on preferences and urgency, revealing insights into collaborative decision-making (AU) | |
| FAPESP's process: | 21/07050-0 - A motivation-based incremental learning framework for robotics |
| Grantee: | Letícia Mara Berto |
| Support Opportunities: | Scholarships in Brazil - Doctorate |