Robotic agents are required to accomplish increasingly com�plex and longer-horizon tasks autonomously. This requires developing novel approaches for computing increasingly elaborate and robust plans that optimize the agents’ behav�ior and allow them to deal with unexpected events. Effective solution approaches for such settings need to manage a rich coupling between three levels of abstraction - task, motion, and control. At the highest level, there is a need to find a policy that prescribes high-level actions that need to be per�formed to achieve some abstract task. For example, a high�level plan may require picking up a tool needed to achieve the goal of fixing a faulty machine. The next level of abstrac�tion is responsible for planning the motions, i.e., physical movements, of the robot that are required for executing each prescribed high-level action. This means, for example, find�ing a sequence of motions that reach a robot configuration from which the tool can be grasped. Finally, at the lowest level, there is a need to control and monitor the execution of each planned motion in order to reach the robot position, or robot configuration, prescribed by the motion planner. Effectively integrating these three components has been established as a challenging sequential decision-making problem that requires integrating skills and tools from dif�ferent research disciplines, investigated by different research communities. This makes the integration of motions and high-level actions especially challenging and constitutes one of the major bottlenecks towards enabling a variety of appli�cations, including mobile manipulation, household robotics, healthcare robotics, and robots for disaster recovery.

Our bridge program aims to bring together re�searchers from different research communities and help cat�alyze the next generation of research in combining AI, machine learning and robotics and developing robots that are capable and efficient at all levels of deliberation and decision-making. We aim to introduce participants coming from different disciplines to a variety of methods of cognitive systems and robotics, thus creating a bridge between robotics and AI and setting the path to the creation of novel methods that combine the tools and expertise from each discipline.

Program Outline Our novel bridge program will offer challenge problems, tu�torials, laser talks and panels on major elements of TMP and learning for TMP that are required to develop capable and dexterous autonomous robotic systems. The content will center around various related themes including motion plan�ning, task planning, robust execution and control, percep�tion, and manipulation, planning under uncertainty and risk, imitation and reinforcement learning, and more.

We will offer a two-day program. Each day will start with two lectures given by prominent researchers, includ�ing speakers who perform research at the crossover between learning, robotics, and AI. It will also include one or two vision and challenge talks, to set the context for the field. Because lectures alone are not enough to have a lasting impact, we will complement them with two hands-on lab sessions each day. During the labs, participants will imple�ment ideas discussed in the talks and will solve problems in a simulated robotic setting. To encourage exchange of ideas, work will be done in groups, with participants from different backgrounds. At the end of each lab session, we will facili�tate a discussion on challenges that were encountered during the labs and potential solutions approaches. To further foster discussions, participants will be asked to provide 1-2 minute laser talks and will be encouraged to bring posters, which will be available throughout the day, with dedicated poster sessions during the breaks.

Please see the AAAI_25_Bridge_TMP-Description file for full details.