No Plan but Everything Under Control: Robustly Solving Sequential Tasks with Dynamically Composed Gradient Descent

2025-03-03

AI Summary

Classical planning paradigms struggle in dynamic, uncertain environments where sequential task solving demands rigid, precomputed plans—limiting adaptability and real-time responsiveness. Method: We propose a planning-free gradient descent framework that implicitly encodes subgoals and task structure via environment feedback-driven dynamic potential field reconstruction. This enables end-to-end, interaction-aware online optimization, integrating feedback-based gradient updates, online parameter adaptation, and physics-informed interaction learning. Contribution/Results: Our approach is the first to naturally emergently exhibit error recovery and hierarchical task decomposition within pure gradient-based optimization—without explicit symbolic planning or discrete state abstraction. It achieves 100% success on the full Blocks World benchmark suite and exceeds 92% success rate over 100 trials on real-robot drawer manipulation tasks—outperforming classical planners in both robustness and efficiency, while reducing computational overhead by an order of magnitude.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Planning, Routing, and Scheduling: Learning for Planning and Scheduling
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

We introduce a novel gradient-based approach for solving sequential tasks by dynamically adjusting the underlying myopic potential field in response to feedback and the world's regularities. This adjustment implicitly considers subgoals encoded in these regularities, enabling the solution of long sequential tasks, as demonstrated by solving the traditional planning domain of Blocks World - without any planning. Unlike conventional planning methods, our feedback-driven approach adapts to uncertain and dynamic environments, as demonstrated by one hundred real-world trials involving drawer manipulation. These experiments highlight the robustness of our method compared to planning and show how interactive perception and error recovery naturally emerge from gradient descent without explicitly implementing them. This offers a computationally efficient alternative to planning for a variety of sequential tasks, while aligning with observations on biological problem-solving strategies.

Problem

• Dynamic gradient descent for sequential tasks
• Robust adaptation to uncertain environments
• Efficient alternative to traditional planning methods

Innovation

• Dynamic gradient descent adjusts potential fields.
• Feedback-driven approach adapts to dynamic environments.
• Computationally efficient alternative to traditional planning.