Experimental laboratory work involves repetitive, labour-intensive manual procedures and strict safety protocols, making it an attractive area for Robotics and Artificial Intelligence (RAI). With research currently focused on technical innovation, there has been little work investigating the human factors of chemist-robot collaboration, and how these technologies will change the nature of work in chemistry. Understanding chemists’ requirements and perspectives is vital to the successful deployment and acceptance of RAI in laboratories. We examined the potential for robot-assisted chemistry experimentation through the perspective of future users by conducting contextual inquiries and semi-structured interviews with nine experimental chemists at various career levels with and without laboratory automation experience. Reflexive thematic analysis was used to gain a situated understanding of how their work could benefit from automation, as well as their perspectives and opinions on working alongside robots. We found that our participants are open to robot-assisted experimentation and have clear ideas for human-robot partnerships in the lab. We identified eight themes characterising chemists’ desired applications, the implementation requirements, and the perspectives on the adoption of robot-assisted experimentation. These findings show that successful acceptance of robot-assisted experimentation depends not only on technical innovation, but also on addressing the practical barriers and concerns of the chemists who will work alongside these systems. The key concern being that RAI should not constrain, but rather complement the role of the chemist in knowledge discovery. The findings contribute to a set of initial design considerations covering hardware, software, and interaction design for successful integration of RAI in experimental laboratory work.
To mitigate the flight safety risks and ensure the feasibility of the trajectories for Unmanned Aerial Vehicles (UAVs) in complex and hazardous environments, this paper proposes a multi-constraint weighted-cost optimization method using an Improved Harris Hawks Optimization (IHHO) algorithm. First, a weighted-cost function is formulated by aggregating path length, flight altitude, flight attitude, threat distance, and the metric of obstacle avoidance response time into a single objective, subject to the kinematic and safety constraints of the UAV. Second, to enhance the optimization performance in path planning, three improvement strategies are introduced to the standard HHO: (1) position updates inspired by natural exploration and trap avoidance mechanisms to augment population diversity and prevent local optima stagnation; (2) a cosine-exponential based non-linear energy decay strategy to replace the original linear decay, enhancing the global search capability; and (3) modulation of the number of Lévy flight iterations to refine local solution accuracy. Finally, experiments on a simulated map demonstrate that the proposed IHHO algorithm efficiently generates optimal trajectories that minimize the weighted cost while harmonizing flight efficiency, path smoothness, and obstacle avoidance safety. This provides a reliable methodological foundation for UAV mission execution in complex environments.
To address the problems of low path planning efficiency and excessive redundant nodes in tomato harvesting manipulators, an improved RRT-APF path planning algorithm is proposed. First, a goal-biased strategy is introduced into the standard Rapidly-exploring Random Tree (RRT) framework to reduce search randomness and improve sampling efficiency. The Artificial Potential Field (APF) method is then integrated with a greedy strategy to perform obstacle avoidance and update nodes. Finally, an improved quadratic B-spline curve is applied to optimize the planned path. In a simple obstacle environment, compared with RRT, RRT*, and APF, the proposed algorithm reduces average planning time by 75.51%, 79.31%, and 52.00%, respectively, and reduces average sampling node count by 88.51%, 85.45%, and 95.78% in 2D scenarios. In 3D scenarios, average planning time is reduced by 99.57%, 99.87%, and 85.71%, and sampling node count is reduced by 98.80%, 98.97%, and 97.23%. In a complex obstacle environment, compared with RRT and RRT*, the algorithm reduces average planning time by 29.41% and 70.73%, and sampling node count by 69.23% and 77.11% in 2D scenarios; in 3D complex environments, average planning time is reduced by 99.80% and 99.86%, and average sampling node count is reduced by 99.49% and 99.38%. Experimental results demonstrate that the improved RRT-APF algorithm has significant advantages in path planning speed and efficiency, providing reliable technical support for Cartesian obstacle avoidance of tomato harvesting manipulators.
Learned locomotion policies for humanoid robots are usually evaluated with all actuators and command channels intact, and how much one-sided degradation of their motor commands they tolerate without retraining is rarely quantified. We addressed this question entirely in physics simulation, using eight independently trained feedforward locomotion policies for the Unitree G1 humanoid model. We attenuated the output-layer weights and biases for all 13 actuated joints on one side of the body by a common factor α, which scales the commanded deviation from the default posture on that side by exactly α while leaving torque limits and joint stiffness unchanged. No retraining was performed. Falls, defined as a tilt of the pelvis (the robot’s root body) exceeding 70° within 5 s, rose steeply as α decreased from about 0.75 to 0.55. In single-rollout sweeps at 0.5 m/s, per-policy thresholds ranged from 65% to 75% remaining output for right-side attenuation and from 60% to 70% for left-side attenuation, and the right side was more fragile in 6 of 8 policies, with none in the reverse direction (exact sign test, p = 0.031). Repeated trials under 10 randomized conditions per policy showed substantial within-policy variability and gave mean 50% fall thresholds of 68.4% (95% CI 66.6%–70.1%) for right-side and 64.6% (62.6%–66.6%) for left-side attenuation (right higher in 7 of 8 policies; Wilcoxon p = 0.039). Across four walking speeds the group-level right-side threshold was never lower than the left, but the per-policy asymmetry reached significance only at 0.5 m/s. At fall onset, joint angles had diverged progressively from the default posture over the preceding ≈0.6 s rather than converging to it. These results provide a reproducible, variability-aware measure of the unilateral command-degradation margin of a learned humanoid locomotion policy. The perturbation was motivated by acute hemiparesis, the sudden weakness of one side of the body after a stroke, before any recovery or relearning, but it is a computational analogy and not a model of human stroke.
Autonomous robot navigation requires path-planning algorithms that maintain reliable performance in environments involving dynamic obstacles, changing goals, and complex terrain. This study extends the evaluation of a previously proposed Modified Rapidly-Exploring Random Tree (Modified RRT) algorithm from two-dimensional benchmark maps to robotic navigation scenarios in three-dimensional simulated environments. A simulation-based robotic navigation framework was developed using ROS 2 and Gazebo. Five scenarios were investigated: reconstruction of selected benchmark maps in 3D simulation, dynamic obstacle avoidance, dynamic goal adaptation, warehouse/campus navigation, and slope traversal. The framework incorporates synchronized RGB and depth sensing with point-cloud generation to characterize simulated environmental observations and support the analysis of perception and navigation performance. The global path planner operates in the horizontal x-y plane, while the robot, obstacles, and terrain are represented in the three-dimensional simulation environment. All four planners were evaluated under a common protocol using a shared costmap, collision checker, stopping budget, and parameter-tuning procedure. Modified RRT achieved success rates of up to 98%, reduced navigation completion time by up to 17.6%, and generated paths 10–15% shorter than standard RRT and 2–10% shorter than the next-best planner, Informed RRT* (IRRT*), at the cost of higher planning time than standard RRT in most scenarios. Under ideal sensing, Modified RRT reduced the near-collision rate by up to 74% in structured environments and 63% in unstructured environments relative to standard RRT. When depth noise, missing measurements, latency, and registration errors were introduced, the reductions were 45–61% and 40–52%, respectively. The results indicate that the Modified RRT sampling mechanisms can improve path-planning and navigation performance across the evaluated simulation scenarios. The findings also show that perception imperfections reduce the near-collision-rate improvements observed under ideal sensing. Overall, this study presents a simulation-based framework for evaluating planar global path planning, robot navigation execution, and simulated multimodal perception in three-dimensional robotic environments.
