Planning Using Belief Summaries for Goal-Directed Manipulation of Articulated Objects with Force and Proprioception
Thavishi Illandara, Michael Hagenow, Julie A. Shah
AI summary
Problem
Manipulating articulated objects to precise joint configurations is challenging due to partial observability, contact-rich dynamics, and the computational intractability of solving continuous POMDPs in real-time. Existing methods often rely heavily on vision or lack robust goal-directed capabilities.
Approach
The authors introduce Planning using Belief Summaries (PuBS), which approximates the continuous POMDP as a low-dimensional MDP by summarizing particle-filter beliefs into just the mean joint state and uncertainty. This tractable MDP is solved offline with reinforcement learning to learn a policy that safely switches between compliant and high-force interaction modes.
Key results
- Novel PuBS framework approximating continuous POMDPs as tractable MDPs via compact belief summaries
- Discrete action space enabling adaptive switching between compliant and high-force interaction modes
- High success rates on the Light Dark POMDP benchmark and real-world articulated object manipulation tasks
- Closed-loop system demonstrating reliable goal-reaching using only force and proprioceptive feedback
Why it matters
Enables robots to safely and accurately manipulate everyday articulated objects in cluttered or vision-obstructed environments without relying on complex visual perception.
Abstract
Enabling robots to manipulate articulated objects is essential for their successful integration into human-centric environments. Such manipulation is often part of a larger multistep task, where achieving a specific joint configuration is necessary for subsequent actions—for example, in a cluttered environment, a cabinet door must be rotated to a precise angle that creates just enough clearance to retrieve an object, beyond which it would collide with surrounding obstacles. In this work, we present an approach to learning goal- directed policies for articulated object manipulation using force and proprioceptive feedback. We formulate the manipulation problem as a Partially Observable Markov Decision Process (POMDP) with a continuous state space and a set of low-level control actions. Due to the limitations of standard POMDP solvers in this setting, we introduce Planning using Belief Summaries (PuBS), which approximates the POMDP as a Markov Decision Process (MDP) over compact particle-filter belief summaries encoding estimated state and uncertainty. This approximate MDP is then solved using reinforcement learning techniques to learn goal-directed policies that enable safe exploration while efficiently guiding the object toward the goal. We evaluate our approach through simulation and real-world robotic experiments, demonstrating reliable goal- reaching performance.