Reinforcement learning (RL)-based quadrotor control policies have achieved impressive performance in tasks such as fast navigation in cluttered environments and drone racing, where the focus is on speed and agility. However, in several applications, such as infrastructure inspection, it is critical to achieve precise, controlled maneuvers with tunable performance. In this article, we present a novel heuristic approach to achieve tunable performance in RL-based Quadrotor control through reward design and termination conditions. We present a novel reward structure containing dual bandwidth exponential functions that achieves a baseline critically damped response in setpoint tracking, with small steady-state error. When trained with a Proximal Policy Optimization (PPO) algorithm, in conjunction with episode truncation conditions, the desired performance is achieved in 6 million time steps of training in a sample-efficient manner. In order to tune the performance about the baseline behavior, we present intuitive heuristic rules to adjust the reward weights and exponential coefficients to achieve faster (acrobatic-like) and slower (inspection-like) settling time performance, while retaining the baseline critically damped response and approximately 2% steady-state error. We evaluate the three RL policies (baseline, acrobatic, and inspection) across 100 trials and show accurate and tunable performance in position and yaw tracking from random initial conditions, thereby demonstrating the effectiveness of the proposed heuristic approach.

This article introduces a novel sample-efficient curriculum learning (CL) approach for training an end-to-end reinforcement learning (RL) policy for robust stabilization of a Quadrotor. The learning objective is to simultaneously stabilize position and yaw-orientation from random initial conditions through direct control over motor RPMs (end-to-end), while adhering to pre-specified transient and steady-state specifications. This objective, relevant in aerial inspection applications, is challenging for conventional one-stage end-to-end RL, which requires substantial computational resources and lengthy training times. To address this challenge, this article draws inspiration from human-inspired curriculum learning and decomposes the learning objective into a three-stage curriculum that incrementally increases task complexity, while transferring knowledge from one stage to the next. In the proposed curriculum, the policy sequentially learns hovering, the coupling between translational and rotational degrees of freedom, and robustness to random non-zero initial velocities, utilizing a custom reward function and episode truncation conditions. The results demonstrate that the proposed CL approach achieves superior performance compared to a policy trained conventionally in one stage, with the same reward function and hyperparameters, while significantly reducing computational resource needs (samples) and convergence time. The CL-trained policy's performance and robustness are thoroughly validated in a simulation engine (Gym-PyBullet-Drones), under random initial conditions, and in an inspection pose-tracking scenario. A video presenting our results is available at https://youtu.be/9wv6T4eezAU.

Efficient transport of construction and demolition waste (CDW) is essential for sustainable urban operations, especially in container-based systems where empty containers are delivered to sites and later collected as a new task when filled. We introduce the Construction and Demolition Waste Collection Problem (CDWCP) as a capacitated multi-vehicle pickup and delivery routing problem and develop a mixed integer linear programming (MILP) formulation that explicitly encodes container lifecycle dynamics in a deterministic single horizon planning. The model represents both delivery (depot → site) and collection (site → depot) tasks, propagates load changes on each task that operates within capacity constraints, and maintains depot flow conservation. The formulation also supports flexible assignment rules, enabling temporally separated delivery and pickup events to be served by the same or different vehicles without altering the core model. A simulation study is conducted on a representative urban scenario show that the MILP produces implementable, cost-effective routes with no capacity violations, improves concurrent vehicle utilization, and yields practical deployable plan. The framework bridges strategic waste-management objectives and day-to-day operational routing for containerized CDW logistics. Video: https://youtu.be/YCcu6uLi4Q4

As robots are increasingly deployed in real-world environments, users expect to issue high-level navigation instructions in natural language and have the robot handle the details. A key challenge is that directly applying Large Language Models (LLMs) to path planning over 3D scene graphs exposes three fundamental failure modes: context windows overflow as map size grows (prompt complexity scales as O(|V|), where |V| is the number of graph nodes), geometric hallucinations produce paths over nonexistent edges, and transmitting scene graphs to cloud APIs risks leaking sensitive spatial data about private environments. We present PathPlanLLM, a neurosymbolic framework that resolves all three limitations by shifting the LLM’s role from planner to data engineer. Rather than serializing the scene graph into the prompt, PathPlanLLM provides the LLM only with a lightweight structural schema, the set of object category labels (typically 85-120 for indoor scenes) and uses a three-stage pipeline: a Parser that maps natural language queries to structured intents, a Semantic Linker that resolves categories to unique node identifiers, and a Generative Code Executor (GCE) that prompts the LLM to synthesize, execute, and iteratively debug Python code against a local, sandboxed scene graph. This architecture achieves O(∣Cl) prompt scaling (sublinear in graph size), privacy by design (no instance-level spatial data ever leaves the local system), and formal path validity (every returned path is executionverified or rejected). We evaluate PathPlanLLM on a 404node hierarchical scene graph across 30 test cases spanning eleven query categories, achieving a 97% success rate and outperforming both baselines by a significant margin using the same Qwen2.5-Coder 14B model.

The process of object search and relocation in an indoor environment, while intuitive for humans, remains a complex challenge for robots. Enabling robots to perform this task autonomously could have a substantial impact towards automation in both domestic and industrial settings. In this article, assuming a familiar environment, a set of target objects with their desired locations, and a robot with limited carrying capacity, we propose a novel methodology for object search and relocation. Given the human-like intuition exhibited by modern large language models (LLMs), they can be leveraged to guide object localization based on environmental context. Our approach integrates LLM-based prediction with graph-based path planning to create a human-like iterative search and relocation framework. The framework consists of an LLM predictor that suggests likely object locations (along with a likelihood score) and an adaptive path planner that dynamically updates the robot’s future path as new information becomes available during the search process. Prior relevant literature that employs LLM inference in indoor environments primarily focuses on assigning new or misplaced objects to appropriate locations. The aspect of enabling a search for a set of missing objects and planning their relocation to desired locations sets this article apart from prior literature. We compare our method to a patrol-based baseline with respect to the distance traversed by the robot in completing the search and relocation mission. In a medium sized indoor environment we demonstrate that it outperforms the baseline on an average by 31.2%.

In this article, we present and experimentally validate a safe docking control strategy designed for an experimental planar floating platform, called the Slider. Three degrees-of-freedom (DOF) platforms like the Slider are used extensively in space industry and academia to emulate micro-gravity conditions on Earth, for validating in-plane Guidance, Navigation and Control (GNC) algorithms. The Slider uses an air cushion (induced by air bearings) to levitate on a smooth flat table, thus emulating the in-plane zero-gravity motion of a spacecraft in orbit. The proposed docking control strategy is applicable in the in-plane approach and docking phases of space docking missions, and is based on the Control Barrier Functions (CBF) approach, where a safe set (a Cardioid), capturing the clearance and direction-of-approach constraints, is rendered positively forward invariant. To enable precise and safe docking in the presence of unmodeled dynamics, disturbances induced by the tether and drifts induced by the non-flat floating surface, we present a switching strategy among the zero and positive level sets of a Cardioid function. In the approach phase, the positive contour of the Cardioid function smoothly steers the Slider platform into the neighborhood of a deadlock point, which is designed to be at a safe distance from the docking port. In the neighborhood of the deadlock point, Slider corrects its proximity and heading until its configuration is well-suited to enter the docking phase. The docking maneuver is initiated by the CBF switching mechanism (positive to zero contour), which expands the safe zone to include the final docking configuration. We present an analysis of the Quadratic program defining the CBF filter, and identify two deadlock points (an asymptotically stable point in the vicinity of the docking port and an unstable point diametrically opposite on the CBF boundary). Both the approach and docking phases are validated through experimentation on the Slider platform, in the presence of tether-induced disturbances and drifts induced by the non-ideal floating surface. In the docking phase, the CBF switching condition effectively handles experimental non-idealities and recovers the slider platform from unsafe configurations. The proposed docking strategy caters to the in-plane (3DOF) approach and docking phases of real space docking missions and is scalable to three-dimensional 6DOF operations, in conjunction with controllers that stabilize the attitude and the out-of-plane degree-of-freedom. Link to the video of experimental demonstration: https://youtu.be/eBiWvnKtG7U?si=QFPD-vm11wydyZSd.

In this article, we propose a control solution for the safe transfer of a quadrotor UAV between two surface robots positioning itself only using the visual features on the surface robots, which enforces safety constraints for precise landing and visual locking, in the presence of modeling uncertainties and external disturbances. The controller handles the ascending and descending phases of the navigation using a visual locking control barrier function (VCBF) and a parametrizable switching descending CBF (DCBF) respectively, eliminating the need for an external planner. The control scheme has a backstepping approach for the position controller with the CBF filter acting on the position kinematics to produce a filtered virtual velocity control input, which an adaptive controller tracks to overcome modeling uncertainties and external disturbances. The experimental validation is carried out with a UAV that navigates from the base to the target using an RGB camera.

In this article, we present a reactive multi-agent coordination architecture for the management of material flows between production/pickup stages and delivery/drop-off stages, in scenarios such as underground mines and automated factory floors. The pickup and delivery stages are modelled as variable task queues, with no a priori information about the inflow into the production queues. The proposed solution coordinates the movement of a group of mobile agents operating between the two stages in a reactive and scalable manner, so that the material is transported from multiple production queues to multiple delivery queues in a balanced/equalized manner. In such a scenario, centralized planners suffer from low reactivity and poor scaling, as the number of agents and number of queues increases. To overcome this problem, we propose a decentralized approach comprising of two separate auction-based task distribution systems for the production and delivery stages, along with behavior-tree based management of agent autonomy and task bidding. Each auction system tracks the length of production/delivery queues and solves the optimal task assignment, based on the bids submitted by the agents. The agents participate in one of the two auction systems at any given time, based on the status of the behavior tree executing the two-stage tasks. We analytically show that the proposed decentralized auctioning approach along with agent autonomy and bidding managed by behavior trees, offers better scalability and reactiveness compared to the centralized approach. The proposed methodology is experimentally validated in a lab environment, in three illustrative material flow management scenarios, using TurtleBot3 robots as agents.