AI and robotics
At CMU-Africa, artificial intelligence (AI) provides a foundation for engineers to build systems that can perceive, plan, and interact with the world to achieve design goals. Faculty and students are working to obtain a deeper understanding of complex systems and leverage sustainable AI-based solutions, such as robotics, for assisting people and improving productivity. In many cases, data analytics play an important role in understanding these systems and machine learning enables predictive analytics which can serve to automate processes, support decision-making, and inform policymakers.
AI and robotics research at CMU-Africa
CMU-Africa aims to use AI and robotic systems to address important problems in areas such as transportation, building systems, manufacturing, energy, agriculture, security, health, and climate. Many projects fall into the AI for Social Good (AI4SG) thematic area and involve working with various publicly available datasets and connecting with specific internal datasets. The goal is to improve our understanding of real-world systems and address specific challenges that can have a positive impact on both society and the environment. For example, the solutions being enabled by AI can reduce time, improve efficiency, increase accuracy, and offer scalability. An emerging thematic area, known as AI ethics, is helping to increase societal awareness of the dangers of some untested AI solutions and the need to carefully balance reward and risk.
AI and robotics research areas
Culturally sensitive social robotics for Africa
CSSR4Africa is a three-year research project to develop culturally sensitive social robotics for Africa. It has the following objectives.
- Identify the verbal and non-verbal social and cultural norms of human interaction that are prevalent in countries in Africa.
- Encapsulate them in the behavioral patterns of social robots so that they can engage with African people in a manner that is consistent with their expectations of acceptable social interaction.
- Demonstrate these culturally-sensitive social robot behaviours in two use cases: one for giving a tour of a university laboratory, and one for assisting and giving directions to visitors at the reception of a university.
Semantically-modulated episodic-procedural memory for cognitive robotics
One of the challenges of AI is to understand how machines can exhibit flexible, cognitive behavior of humans. One approach—the situation model framework—involves behavioral episodes, recombinable models that link a current state, an action, and a goal state. Behavioral episodes can be internally simulated to determine the effects of carrying out actions in the current context before performing the action. The goal of this project is to develop a computational model of behavioral episodic memory using contemporary deep neural networks in the form of a semantically-modulated joint episodic-procedural associative memory, allowing high-level language descriptions of a desired action to condition the recall of appropriate behavioral episodes.
Pictured: Joint episodic-procedural memory as an explicit network of associations between perceptions and actions, drawn from episodic and procedural memories, unwinding the temporal and causal relationships between specific perceptions and actions that are implicit in the mappings of other perceptuo-motor representations
Facially-expressive robots for human-robot interaction
Human-robot interaction in the African context has motivated researchers to revisit the uncanny valley challenge in robotics—avoiding a negative emotional response towards robots that bear a close, but not sufficiently close, resemblance to humans, especially when the robot alters its facial expressions—and how different cultures may have different expectations of what makes a robot look and behave "human." In the AI-robotics course Humanoid Robotic and Cognition, students are taught to build a replica of the open-source robot Eva to learn the technical challenges of synchronizing lip and facial gestures with audio output.
Enabling AI models to understand physics
CMU-Africa is working to enable AI models to understand physics, which plays a critical role in how robots engage with their environments and with people. At a glance, humans can assess whether a floor surface appears slippery or safe to walk on, if a car tire seems properly inflated or deflated, or if a package is securely sealed or at risk of opening during transit. However, the understanding of the computational framework enabling machines to swiftly process and interpret such information remains limited. To ensure the deployment of safe and beneficial systems in practical scenarios, it's crucial for AI/ML models to encompass both our intuitive and explicit grasp of physics.
Visualization of learned simulator predictions (Left: Particle-based physics datasets. Right: Mesh-based cylinder flow dataset)
Get involved
- Learn more about CMU-Africa's Master of Science in Engineering Artificial Intelligence
- Contact Timothy Brown if you are interested in collaborating with CMU-Africa
Master of Science in Engineering Artificial Intelligence
This 16-20 month program opens the door to advanced skills that will enable engineers to design powerful solutions to today's challenges.
Explore Other Research Topics