An official website of the United States government
Programmable photonic circuits of reconfigurable interferometers can be used to implement arbitrary operations on optical modes, providing a flexible platform for accelerating tasks in quantum simulation, signal processing, and artificial intelligence. A major obstacle to scaling up these systems is static fabrication error, where small component errors within each device accrue to produce significant errors within the circuit computation. Mitigating this error usually requires numerical optimization dependent on real-time feedback from the circuit, which can greatly limit the scalability of the hardware. Here we present a deterministic approach to correcting circuit errors by locally correcting hardware errors within individual optical gates. We apply our approach to simulations of large scale optical neural networks and infinite impulse response filters implemented in programmable photonics, finding that they remain resilient to component error well beyond modern day process tolerances. Our results highlight a potential way to scale up programmable photonics to hundreds of modes with current fabrication processes.
more »
« less
Task-Level Control and Poincaré Map-Based Sim-to-Real Transfer for Effective Command Following of Quadrupedal Trot Gait
The ability of quadrupedal robots to follow commanded velocities is important for navigating in constrained environments such as homes and warehouses. This paper presents a simple, scalable approach to realize high fidelity speed regulation and demonstrates its efficacy on a quadrupedal robot. Using analytical inverse kinematics and gravity compensation, a task-level controller calculates joint torques based on the prescribed motion of the torso. Due to filtering and feedback gains in this controller, there is an error in tracking the velocity. To ensure scalability, these errors are corrected at the time scale of a step using a Poincar´e map (a mapping of states and control between consecutive steps). A data-driven approach is used to identify a decoupled Poincar´e map, and to correct for the tracking error in simulation. However, due to model imperfections, the simulation-derived Poincar´e map-based controller leads to tracking errors on hardware. Three modeling approaches – a polynomial, a Gaussian process, and a neural network – are used to identify a correction to the simulation-based Poincar´e map and to reduce the tracking error on hardware. The advantages of our approach are the computational simplicity of the task-level controller (uses analytical computations and avoids numerical searches) and scalability of the sim-to-real transfer (use of low-dimensional Poincar´e map for sim-to-real transfer). A video is in this shortened link: http://tiny.cc/humanoids23
more »
« less
- Award ID(s):
- 2128568
- PAR ID:
- 10540512
- Publisher / Repository:
- IEEE
- Date Published:
- ISBN:
- 979-8-3503-0327-8
- Page Range / eLocation ID:
- 1 to 8
- Format(s):
- Medium: X
- Location:
- Austin, TX, USA
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
We introduce Spatial Predictive Control (SPC), a technique for solving the following problem: given a collection of robotic agents with black-box positional low-level controllers (PLLCs) and a mission-specific distributed cost function, how can a distributed controller achieve and maintain cost-function minimization without a plant model and only positional observations of the environment? Our fully distributed SPC controller is based strictly on the position of the agent itself and on those of its neighboring agents. This information is used in every time step to compute the gradient of the cost function and to perform a spatial look-ahead to predict the best next target position for the PLLC. Using a simulation environment, we show that SPC outperforms Potential Field Controllers, a related class of controllers, on the drone flocking problem. We also show that SPC works on real hardware, and is therefore able to cope with the potential sim-to-real transfer gap. We demonstrate its performance using as many as 16 Crazyflie 2.1 drones in a number of scenarios, including obstacle avoidance.more » « less
-
Simulation parameter settings such as contact models and object geometry approximations are critical to training robust robotic policies capable of transferring from simulation to real-world deployment. Previous approaches typically handcraft distributions over such parameters (domain randomization), or identify parameters that best match the dynamics of the real environment (system identification). However, there is often an irreducible gap between simulation and reality: attempting to match the dynamics between simulation and reality across all states and tasks may be infeasible and may not lead to policies that perform well in reality for a specific task. Addressing this issue, we propose AdaptSim, a new task-driven adaptation framework for sim-to-real transfer that aims to optimize task performance in target (real) environments -- instead of matching dynamics between simulation and reality. First, we meta-learn an adaptation policy in simulation using reinforcement learning for adjusting the simulation parameter distribution based on the current policy's performance in a target environment. We then perform iterative real-world adaptation by inferring new simulation parameter distributions for policy training, using a small amount of real data. We perform experiments in three robotic tasks: (1) swing-up of linearized double pendulum, (2) dynamic table-top pushing of a bottle, and (3) dynamic scooping of food pieces with a spatula. Our extensive simulation and hardware experiments demonstrate AdaptSim achieving 1-3x asymptotic performance and ∼2x real data efficiency when adapting to different environments, compared to methods based on Sys-ID and directly training the task policy in target environments.more » « less
-
The problems of changing the walking speed and stride length of impact-free gaits for point-foot planar bipeds are addressed. The impact-free gaits are designed using an approach developed in prior work. It is shown that the impulse controlled Poincar´e map (ICPM) approach can be modified to transition between orbits defining gaits with different walking speeds, and the continuous controller can be changed during the swing phase to transition between gaits that have distinct stride lengths. The effectiveness of the approaches is demonstrated using simulations carried out on a five-link biped.more » « less
-
Experimental and Analytical Prescribed-Time Trajectory Tracking Control of a 7-DOF Robot ManipulatorWe present an analytical design and experimental verification of trajectory tracking control of a 7-DOF robot manipulator, which achieves convergence of all tracking errors to the origin within a finite terminal time. A key feature of this control strategy is that this terminal convergence time is explicitly prescribed by the control designer, and is thus independent of the initial conditions of the tracking errors. In order to achieve this beneficial property of the proposed controller, a scaling of the state by a function of time that grows unbounded towards the terminal time is employed. Through Lyapunov analysis, we first demonstrate that the proposed controller achieves regulation of all tracking errors within the prescribed time as well as the uniform boundedness of the joint torques, even in the presence of a matched, non-vanishing disturbance. Then, through both simulation and experiment, we demonstrate that the proposed controller is capable of converging to the desired trajectory within the prescribed time, despite large initial conditions of the tracking errors and a sinusoidal disturbance being applied in each joint.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1109/Humanoids57100.2023.10375183
