Planning is fundamental to autonomous robots, enabling purposeful decision-making, adaptation to change, and sustained, reliable operation over long time horizons. Yet planning remains fragmented across the AI and control communities, which often pursue parallel solutions with limited interaction. AI-driven planning (e.g., RL) offers adaptability and reasoning over complex information but struggles to guarantee resilient and reliable long-term behavior, where performance must be maintained under disturbances and remain predictable across extended deployments. Control-driven planning (e.g., MPC) provides structure, predictability, and robustness, but is limited in flexibility, scalability, and its ability to integrate heterogeneous knowledge. Neither approach alone fully supports autonomy that must persist, evolve, and remain trustworthy over extended deployments. 

This workshop aims to bridge these perspectives by promoting a shared view of planning for long-term robot autonomy. Unlike workshops focused on specific domains or isolated aspects such as perception or safety, this event treats planning as a foundational, cross-cutting problem. Its emphasis is on integrating AI-driven reasoning with control-driven structure to achieve resilient and reliable long-horizon behavior. 

Key research questions include: what are the benefits and limitations of integrating control and AI for robotic planning? Can we provide shared representations for goal, constraints and uncertainty to bridge the gap between AI and control? Which approaches can capture long-horizon robustness under distribution shift, rare events, and cumulative degradation? 

Through invited talks, contributed presentations, and an interactive panel, the workshop seeks to define a common research agenda and foster collaboration toward autonomous systems that plan intelligently, reliably, and with sustained resiliency. 

Participants will gain insights into: 

• Shared abstractions: Which representations—such as temporal logic (LTL/STL), constraint graphs, Lyapunov certificates, or value‑function/cost‑to‑go formulations—can jointly express goals, constraints, and uncertainty in ways compatible across AI and control, enabling planners to remain reliable as tasks and environments evolve?

• Hybrid adaptability–predictability: How can AI‑driven adaptability be combined with control‑driven structure to preserve long‑horizon coherence? Relevant families include learning‑based MPC, safe RL with control barrier functions, hierarchical RL–MPC schemes, optimization‑based TAMP, and neural or diffusion planners equipped with stability filters, with the goal of avoiding brittle or drifting policies.

• Preventing long‑term drift: Which interaction mechanisms between AI reasoning and control structure—such as safety shields, certified policy updates, robust MPC, or adaptive controllers using learned uncertainty estimates—can keep planners aligned with safe and intended behavior over time?

• Learning with stability: How can experience‑driven improvement be aligned with stability requirements so that learning enhances performance without eroding reliability? Directions include stability‑aware policy learning, adaptive control with learned dynamics, and hybrid symbolic–continuous planners that integrate data‑driven components without compromising guarantees.

• Evaluation for long‑horizon autonomy: Which evaluation practices meaningfully capture long‑term robustness and sustained autonomy—e.g., stress‑testing under distribution shift, temporal‑consistency metrics, or benchmarks measuring cumulative degradation or drift—to assess whether planners remain reliable in realistic deployment conditions?