AKS

Autonomous and Kooperative Systems Lab

Institute of Technology, University of Tartu

We are affiliated with Institute of Technology, University of Tartu. Our core aim is to develop algorithms that allow robots to perform complex deision making under real-time systems. The application domain of our research spans across mobile robot navigation, and autonomous driving to manipulation.

Latest News and Information

3 papers (including 2 RAL transfer) and 1 poster accepted at IROS 2025
1 Acceptance in IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING
2 Papers accepted at ICRA 2025.
4 papers accepted at IROS 2024.

Paper accepted at IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING 2025

MMD-OPT: Maximum Mean Discrepancy-Based Sample Efficient Collision Risk Minimization for Autonomous Driving

This paper shows that MMD-OPT, a sample-efficient method leveraging Maximum Mean Discrepancy (MMD) in a Reproducing Kernel Hilbert Space, can effectively minimize collision risk under arbitrary obstacle prediction distributions. It demonstrates through simulations that MMD-OPT achieves safer trajectories with fewer samples compared to CVaR-based approaches.

Paper accepted at Robotics and Automation Letters (RAL) and to be presented at IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2025

π-MPPI: A Projection-Based Model Predictive Path Integral Scheme for Smooth Optimal Control of Fixed-Wing Aerial Vehicles

This paper introduces π-MPPI, a Model Predictive Path Integral control framework that integrates a projection filter π to enforce smoothness by bounding control magnitudes and higher-order derivatives. Applied to fixed-wing vehicles, π-MPPI achieves smoother and more robust performance than standard MPPI with minimal computational overhead.

Paper accepted at IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2025

Diffusion-FS: Multimodal Free-Space Prediction via Diffusion for Autonomous Driving

This paper proposes a new method to predict drivable corridors — safe paths within the road — using only a single front-view camera. It introduces ContourDiff, a diffusion-based model that learns to generate and refine corridor contours in a self-supervised way using the car’s future trajectories, achieving accurate and interpretable free-space predictions on both nuScenes and CARLA datasets.

Paper accepted at International Conference on Control & Automation (ICCA) 2025

DISCO: Diffusion-based Inter-Agent Swarm Collision-free Optimization for UAVs

This paper presents a diffusion-based generative model for coordinated trajectory planning in multi-UAV swarms, using Bernstein polynomial coefficients and self-attention layers to produce diverse and feasible motion plans. Trained on expert demonstrations and equipped with a safety filter for collision avoidance, the method achieves high success rates in generating smooth, collision-free trajectories for large UAV swarms.

Paper accepted at IEEE International Conference on Robotics and Automation (ICRA) 2025

DA-VIL: Adaptive Dual-Arm Manipulation with Reinforcement Learning and Variable Impedance Control

This paper presents a novel dual-arm manipulation pipeline that integrates policy learning with gradient-based optimization to adaptively tune controller gains for dynamic impedance modulation. Evaluated on trajectory-tracking tasks with diverse objects, the approach demonstrates superior stability and dexterity compared to three established dual-arm control methods.

Paper presented at IEEE International Conference on Robotics and Automation (ICRA) 2025

CrowdSurfer: Sampling Optimization Augmented with Vector-Quantized Variational AutoEncoder for Dense Crowd Navigation

This paper presents a crowd navigation method that combines generative modeling with inference-time optimization to produce long-horizon local plans in dynamic and densely populated environments. By using a Vector Quantized Variational Autoencoder to learn a trajectory prior and refine it through sampling-based optimization, the approach achieves a 40% higher success rate and 6% faster travel time compared to DRL-VO, without requiring explicit dynamic obstacle prediction.

Paper accepted at Robotics and Automation Letters and presented at IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2024

PRIEST: Projection guided sampling-based optimization for autonomous navigation

This paper presents a novel gradient-free trajectory optimizer that explores multiple homotopies using a projection-guided sampling strategy to generate feasible, high-quality paths in unknown and dynamic environments. Integrated within the ROS navigation stack, the method improves success rates by 7–13% and travel efficiency by up to 2× compared to existing planners, outperforming MPPI, CEM, and VPSTO on standard benchmarks.

Paper accepted at Robotics and Automation Letters and presented at IEEE International Conference on Robotics and Automation (ICRA) 2024

VACNA: Visibility-Aware Cooperative Navigation with Application in Inventory Management

The paper introduces visibiity-aware cooperative navigation between a UAV and a UGV with applications in inventory management. We develop a novel sampling based optimizer that can compute real-time coordination trajectories while satisfying occlusion and collision constraints

2 Papers presented at IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2023

End-To-End Learning of Behavioural Inputs for Autonomous Driving in Dense Traffic

Hilbert Space Embedding-Based Trajectory Optimization for Multi-Modal Uncertain Obstacle Trajectory Prediction

First paper proposes a new differentiable trajectory optimizer and embeds it into an end-to-end learning pipelines for autonomous driving.

Second paper proposes a chance constrained optimizer to handle multi-modal obstacle trajecory predictions leveraging the concept of embedding distribution in Reproducing Kernel Hilbert Space.

Paper presented at IEEE International Conference on Robotics and Automation (ICRA) 2023

AMSwarm: An Alternating Minimisation Approach for safe motion planning of Quadrotor swarms in cluttered and Dynamic Environments.

This paper proposes an online and scalable algorithm capable of generating safe and kinematically feasible trajectories for quadrotor swarms.

3 Papers presented at IEEE Conference on Automation Science and Engineering (CASE) 2023

Bi-Level Optimization Augmented with Conditional Variational Autoencoder for Autonomous Driving in Dense Traffic

UAP-BEV: Uncertainty Aware Planning Using Bird's Eye View Generated from Surround Monocular Images

Disentangling Planning and Control for Non-Prehensile Tabletop Manipulation

First paper proposes simultaneous optimization of behavioral and motion planning layers of autonomous driving

Second paper proposes a chance-constrained optimizer that can take plan safe trajectories for autonomous cars based on Birds Eye View prediction of the occupancy map.

Third paper develops a lower-level RL based controller for pushing objects with a manipulator and couples it with a high-level planner like A* for object manipulation.

2 Papers accepted at American Control Conference (ACC) 2023

UrbanFly: Uncertainty-Aware Planning for Navigation Amongst High-Rises with Monocular Visual-Inertial SLAM Maps

GPU Accelerated Batch Trajectory Optimization for Autonomous Navigation

First paper proposes uncertainty aware planning for monocular slam systems

Second paper proposes a vectorized batch optimizer to solve tens of hundreds of trajectory optimization in parallel over GPUs.

Page updated

Google Sites

Report abuse