Daniel Rakita

Haoxiang You, Yilang Liu, Davis Zong et al.

We present the stochastic decoupled policy gradient (SDPG), a lightweight visual reinforcement learning (RL) method that trains diverse visuomotor control policies end-to-end within a few hours on a single NVIDIA RTX 4080 GPU. SDPG estimates policy gradients via random perturbations of trajectory rollouts, requiring orders of magnitude fewer batch-rendered environments and substantially reducing compute and memory overhead. On visual MuJoCo benchmarks, SDPG consistently outperforms baseline methods in training time, memory usage, and rewards. Finally, to support future research, we introduce a suite of realistic visual robotics benchmarks spanning dexterous manipulation, challenging locomotion, and demonstrate effective sim-to-real transfer on physical hardware.

Chen Liang, Xiatao Sun, Qian Wang et al.

Zeroth-Order (ZO) optimization is pivotal for scenarios where backpropagation is unavailable, such as memory-constrained on-device learning and black-box optimization. However, existing methods face a stark trade-off: they are either sample-inefficient (e.g., standard finite differences) or suffer from high variance due to randomized estimation (e.g., random subspace methods). In this work, we propose Coherent Coordinate Descent (CoCD), a deterministic, sample-efficient, and budget-aware ZO optimizer. Theoretically, we formalize the notion of gradient coherence and demonstrate that CoCD is equivalent to Block Cyclic Coordinate Descent (BCCD) with ``warm starts,'' effectively converting historical (stale) gradients from a liability into a computational asset. This mechanism enables $O(1)$ query complexity per step while maintaining global descent directions. Furthermore, we derive error bounds revealing a counter-intuitive insight: larger finite-difference step sizes can induce an implicit smoothing effect on the optimization landscape by reducing the effective smoothness constant, thereby improving convergence stability. Experiments on MLP, CNN, and ResNet architectures (up to 270k parameters) demonstrate that CoCD significantly outperforms BCCD in terms of sample efficiency and convergence loss/accuracy, and exhibits superior stability over randomized ZO methods. Our results suggest that deterministic, structure-aware updates offer a superior alternative to randomization for lightweight ZO optimization.

Constantinos Chamzas, Carlos Quintero-Peña, Zachary Kingston et al.

Recently, there has been a wealth of development in motion planning for robotic manipulation new motion planners are continuously proposed, each with their own unique strengths and weaknesses. However, evaluating new planners is challenging and researchers often create their own ad-hoc problems for benchmarking, which is time-consuming, prone to bias, and does not directly compare against other state-of-the-art planners. We present MotionBenchMaker, an open-source tool to generate benchmarking datasets for realistic robot manipulation problems. MotionBenchMaker is designed to be an extensible, easy-to-use tool that allows users to both generate datasets and benchmark them by comparing motion planning algorithms. Empirically, we show the benefit of using MotionBenchMaker as a tool to procedurally generate datasets which helps in the fair evaluation of planners. We also present a suite of 40 prefabricated datasets, with 5 different commonly used robots in 8 environments, to serve as a common ground to accelerate motion planning research.

Daniel Rakita, Bilge Mutlu, Michael Gleicher

In this paper, we present a meta-algorithm intended to accelerate many existing path optimization algorithms. The central idea of our work is to strategically break up a waypoint path into consecutive groupings called "pods," then optimize over various pods concurrently using parallel processing. Each pod is assigned a color, either blue or red, and the path is divided in such a way that adjacent pods of the same color have an appropriate buffer of the opposite color between them, reducing the risk of interference between concurrent computations. We present a path splitting algorithm to create blue and red pod groupings and detail steps for a meta-algorithm that optimizes over these pods in parallel. We assessed how our method works on a testbed of simulated path optimization scenarios using various optimization tasks and characterize how it scales with additional threads. We also compared our meta-algorithm on these tasks to other parallelization schemes. Our results show that our method more effectively utilizes concurrency compared to the alternatives, both in terms of speed and optimization quality.

Daniel Rakita, Bilge Mutlu, Michael Gleicher

In this work, we present a novel sampling-based path planning method, called SPRINT. The method finds solutions for high dimensional path planning problems quickly and robustly. Its efficiency comes from minimizing the number of collision check samples. This reduction in sampling relies on heuristics that predict the likelihood that samples will be useful in the search process. Specifically, heuristics (1) prioritize more promising search regions; (2) cull samples from local minima regions; and (3) steer the search away from previously observed collision states. Empirical evaluations show that our method finds shorter or comparable-length solution paths in significantly less time than commonly used methods. We demonstrate that these performance gains can be largely attributed to our approach to achieve sample efficiency.

Daniel Rakita, Haochen Shi, Bilge Mutlu et al.

In this work, we present a per-instant pose optimization method that can generate configurations that achieve specified pose or motion objectives as best as possible over a sequence of solutions, while also simultaneously avoiding collisions with static or dynamic obstacles in the environment. We cast our method as a multi-objective, non-linear constrained optimization-based IK problem where each term in the objective function encodes a particular pose objective. We demonstrate how to effectively incorporate environment collision avoidance as a single term in this multi-objective, optimization-based IK structure, and provide solutions for how to spatially represent and organize external environments such that data can be efficiently passed to a real-time, performance-critical optimization loop. We demonstrate the effectiveness of our method by comparing it to various state-of-the-art methods in a testbed of simulation experiments and discuss the implications of our work based on our results.