SmoothTurn: Learning to Turn Smoothly for Agile Navigation with Quadrupedal Robots

Jul 16, 2026

Quadrupedal robots have strong real-world value in time-critical missions such as fire rescue and industrial inspection, where navigation must remain agile and high-speed. Traditional goal-conditioned methods usually learn single-goal reaching and often stop or slow before switching targets, preventing anticipation and momentum transfer across direction changes. SmoothTurn reformulates agile navigation as sequential local navigation and learns a rapid, smooth-turning control policy via a sequential goal-reaching reward, lookahead-based observations, and an automatic goal curriculum.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

Introduction

Motivation and challenges

Limitations of existing goal-reaching methods

Sequential local navigation formulation

SmoothTurn approach

Sequential goal-reaching reward

Lookahead observation design

Automatic goal curriculum

Evaluation and deployment

SmoothTurn: Learning to Turn Smoothly for Agile Navigation with Quadrupedal Robots Zunzhi You, Yunke Wang, Haolan Guo, Chang Xu arXiv :2603 . 12842v2 [ cs .RO] 12 Jul 2026 Abstract—Quadrupedal robots show great potential for valuable real-world applications such as fire rescue and industrial inspection. Such applications often require urgency and the ability to navigate agilely, which in turn demands the capability to change directions smoothly while running at high speed. Existing approaches for agile navigation typically learn a single-goal reaching policy by encouraging the robot to stay at the target position after reaching it. As a result, when the policy is used to reach sequential goals that require changing directions, it cannot anticipate upcoming maneuvers or maintain momentum across the switch of goals, thereby preventing the robot from fully exploiting its agility potential. In this work, we formulate the task as sequential local navigation, extending the single-goal-conditioned local navigation formulation in prior work. We then introduce SmoothTurn, a learning-based control framework that learns to turn smoothly while running rapidly for agile sequential local navigation. The framework adopts a novel sequential goal-reaching reward, an expanded observation space with a lookahead window for future goals, and an automatic goal curriculum that progressively expands the difficulty of sampled goal sequences based on the goal-reaching performance. The trained policy can be directly deployed on real quadrupedal robots with onboard sensors and computation. Both simulation and real-world empirical results show that SmoothTurn learns an agile locomotion policy that performs smooth turning across goals, with emergent behaviors such as controlling momentum when switching goals, facing towards the future goal in advance, and planning efficient paths. I. INTRODUCTION Legged robots can operate in environments where wheeled platforms often struggle, including stairs, debris, discontinuous footholds, and uneven outdoor terrain. Quadrupedal robots are especially useful because they provide stable support and can traverse complex terrain while carrying onboard sensing and computation [1] . Many target applications of quadrupedal robots are time-sensitive. For example, in fire rescue [2], rapid reconnaissance and response can reduce risk to firefighters and save more lives and property. Even in routine service scenarios such as hospital delivery or campus logistics, efficient navigation through sequences of waypointsis highly desirable. Achieving high speed in the real world requires more than fast straight-line locomotion. Navigation in buildings and cluttered spaces involves frequent direction changes to follow corridors, pass doorways, and maneuver around corners. These maneuvers can be viewed as reaching a sequence of goals. For example, when leaving a room through a doorway, one needs to first reach the doorway (the initial goal), and then change the direction to navigate to a location outside the *All authors are with School of Computer Science, University of Sydney. Fig. 1. Composited images of SmoothTurn deployed on a Unitree Go2 performing agile navigation in an indoor office environment. The learned policy enables the robot to maintain momentum and high speed while executing turns rapidly through corridors and corners. room (the subsequent goal) . A stop-and-go behavior at each turn greatly increases overall traversal time, while turning too aggressively without anticipation risks instability and falls. To date, most legged locomotion controllers are designed to track commanded base velocities [3]–[5] . This approach is effective for producing robust gaits, but it places the responsibility for when and how much to turn on an operator or an upstream navigation module that must generate velocity andyaw commands. More recently, goal-conditioned approaches learn local navigation by conditioning the policy

FAQs

Why do single-goal reaching policies struggle with sequential agile navigation?

Because the learned behavior tends to slow down or settle at each reached goal, it cannot anticipate upcoming maneuvers or maintain momentum when switching directions between sequential goals.

How does SmoothTurn define the navigation task?

SmoothTurn formulates navigation as sequential local navigation, where the robot receives an ordered sequence of local goals and must reach the previous goal before moving to the next.

What key training components enable SmoothTurn to turn smoothly at high speed?

The method uses a sequential goal-reaching reward to encourage continuous progress, an expanded observation with a multi-goal lookahead window for anticipation, and an automatic goal curriculum that progressively increases sequence difficulty based on performance.

Jul 16, 2026

Upload by Lucas Martin