JPL designs and builds systems that achieve incredibly ambitious goals, as evidenced by the Curiosity rover traversing on Mars, the highly complex Cassini spacecraft orbiting Saturn, and the compelling concept for retrieving a boulder from an asteroid and inserting it into a lunar orbit to create a nearby target to be investigated by astronauts. Autonomy -- which includes creating and optimizing spacecraft activity plans, executing nominal and critical activities, analyzing and interpreting data, assessing system health and environmental conditions, and diagnosing and responding to off-nominal situations -- is a fundamental part of achieving these goals.
A spacecraft’s ability to sense, plan, decide, and take actions to accomplish science and other mission objectives depends on the flight systems that implement the intended functionality and the operators who command them. Historically, success has depended on an ability to predict the relevant details of remote environments well enough to perform the mission safely and effectively.
As NASA and JPL advance our knowledge frontier, science questions become more sophisticated and mission environments more difficult, harsh, and inaccessible. This leads to new challenges as shown in the following examples:
- Hazardous conditions such as the high radiation levels surrounding interesting destinations like Europa and the toxic atmospheres of planetary bodies like Venus limit mission lifetime and leave multiple complex activities with very few possible ground interventions.
- Concepts for missions to free-floating, active small bodies and other destinations with unconstrained or unknown environments need more sophisticated perception-rich behaviors and flexible in situ decision-making (see also Robotics section).
- Multi-element missions, such as a potential Mars Sample Return, which would involve physical interaction of multiple systems to capture and transfer samples as well as launching off of another celestial body (see also Robotics section).
- Concepts for long-duration missions such as the Kuiper Belt exploration and ambitious interstellar explorers must operate in unknown environments and survive equipment failures over the course of decades.
Because science investigations are expected to deliver increasingly exciting results and discoveries, systems are becoming progressively complex and engineering designs must adapt by improving in situ functionality. For example, these compelling yet challenging investigations will need to revise their operational tactics—and sometimes even their science objectives—“on the fly” in response to unforeseen mission risks, requiring unprecedented onboard decision-making on short timescales. As missions visit more distant and formidable locations, the job of the operations team becomes more challenging, increasing the need for autonomy.
Autonomy capabilities have the potential to advance rapidly, and they must do so to support next-generation space science and exploration goals. Advances in autonomous behaviors and decision-making and related fields that come from the academic, industry, and government arenas are being adapted for space system applications. At the same time, current mission approaches are reaching the limits of what can be accomplished without such advances. JPL currently strives for autonomy technology development, maturation, and infusion in six principal areas:
- Planning, scheduling, and execution
- Robust critical activities
- In situ data interpretation and learning
- State-awareness and system health management
- Perception-rich behaviors (see Robotics section)
- Physical interaction (see Robotics section)
Selected Research Topics
The functionality requirements of science missions will, and must, continue to evolve, yet the need for extreme reliability in flight systems remains a critical factor. In the past, deep-space missions were commanded almost entirely from the ground, with ingenuity and patience overcoming the difficulties of light-time delays. Except during critical events such as entry, descent, and landing on Mars and one-time activities such as orbit insertions, reliability was achieved largely via “safing” responses that used block-level redundancy with fail-over based on straightforward, simplistic system behavior checks. Now that surface missions—with their continuous uncertainties associated with operating on a planetary surface—are an established mission class, meeting science objectives requires real-time, goal-directed, situationally aware decision-making. To meet these needs, technological capabilities are evolving to close more decision loops onboard the spacecraft, both for mission planning and operations and for fault response. Future spacecraft and space missions will rely heavily on the in situ decision-making enabled by designing for autonomy in both hardware- and software-based functionality.
JPL’s autonomous operations capabilities include automated planning, intelligent data understanding, execution of robust critical activities such as entry, descent and landing (EDL), and situational- and self-awareness. These capabilities can be used in both flight and ground systems to support both deep-space and Earth-orbiting missions. Autonomous operations involve a range of automated behaviors for spacecraft including onboard science event detection and response, rapid turnaround of ground science plans, and efficient re-planning and recovery in response to anomalous events. Successes in this area include (1) the use of onboard image analysis to automatically identify and measure high-priority science targets for the rovers on Mars and (2) the use of automated planning onboard an Earth satellite to manage routine science activities and automatically record events such as volcanic eruptions, flooding, and changes to polar ice caps.
Planning, Scheduling, and Execution
An important autonomy capability for current and future spacecraft is onboard decision-making, where spacecraft activity plans are autonomously created and executed, enabling a spacecraft to safely achieve a set of science goals without frequent human intervention. To provide this capability, planning, scheduling, and execution software must be capable of rapidly creating and validating spacecraft plans based on a rich model of spacecraft operations. Plans typically correspond to spacecraft command sequences that are executed onboard and that ensure the spacecraft is operated within safe boundaries. For each spacecraft application, the planning system contains a model of spacecraft operations that describes resource, state, temporal, and other spacecraft operability constraints. This information enables the planning system to predict resource consumption, such as power usage, of variable-duration activities, keep track of available resource levels, and ensure that generated plans do not exceed resource limits. Planning and scheduling capabilities typically include a constraint management system for reasoning about and maintaining constraints during plan generation and execution as well as a number of search strategies for quickly determining valid plans. A graphical interface provides visualizing the plans/schedules to operators on the ground.
Once plans are generated, plan execution can be monitored onboard to ensure plan activities are executed successfully. If unexpected events such as larger-than-predicted power usage or identification of new science goals occur, plans can be dynamically modified to accommodate the new data. To support re-planning capabilities, a planning system monitors the current spacecraft state and the execution status of plan activities. As this information is acquired, future-plan projections are updated. These updates may cause new conflicts and/or opportunities to arise, requiring the planning system to re-plan to accommodate the new data. In order to reason about science goal priorities and other plan quality measures, optimization capabilities can be used to search for a higher quality plan. User-defined preferences can be incorporated and plan quality computed based on how well the plan satisfies these preferences. Plan optimization can also be performed in an iterative fashion, where searches are continually performed for plan modifications that could improve the overall plan score.
Robust Critical Activities
Communication delays and constraints often preclude direct ground-in-the-loop involvement during critical activities such as entry, descent, landing, orbit insertions, proximity operations, and observation of transient phenomena; therefore, JPL’s space assets must rely on onboard control and autonomy. Future missions likely will have more challenging requirements for operating in even more complex and less known space and planetary environments. These include major shrinking of landing ellipses, closer and more precise proximity operations (e.g., touch-and-go sampling maneuvers and flying through and taking samples of vents and geysers), and more complex measurements of transient phenomena. Along with the evolving and sophisticated sensing suite, these demanding requirements call for more capable onboard reasoning and decision-making for critical real-time applications. Capabilities such as terrain relative navigation, terrain hazard assessment for landing, onboard nonlinear state estimation, sensor fusion, real-time optimal guidance laws for trajectory planning with constraints, and coordinated multi-instrument observations require sophisticated onboard computing and reasoning about larger volumes of data with greater uncertainty. JPL is developing novel, cost-effective techniques to mature, validate, and verify these sophisticated system capabilities for infusion into future missions.
In Situ Data Interpretation and Learning
Autonomous capabilities continue to extend the reach of science investigations conducted by remote spacecraft while maintaining system reliability and managing risk. Applications include triaging data for downlink when more data is collected than can be transmitted immediately to Earth and responding to features and events in the remote environment more rapidly than would be possible with the ground in the loop. Past examples include:
- Volcano and flood detection onboard Earth-orbiting spacecraft, enabling rapid follow-up imaging;
- Real-time detection of methane during airborne campaigns, enabling adjustment of the flight path to track the plume and identify and characterize the source;
- Detection of interesting geologic features in rover imagery data and subsequent triggering of the collection of follow-up detail imagery in the same operations sol;
- Dust-devil detection onboard Mars rovers.
Future applications include the use of onboard landmark detection and matching to provide pinpoint landing for Mars surface missions and the use of onboard data understanding and dynamic instrument parameter adjustment to provide rapid event detection and response for missions to primitive bodies.
State-Awareness and System Health Management
In order to accomplish increasingly ambitious exploration goals in unknown environments, the space systems must have sufficient knowledge and capabilities to realize these goals and to identify and respond to off-nominal conditions. Decisions made by a system—or by its operators—are only as good as the quality of knowledge about the state of the system and its environment. In highly complex and increasingly autonomous spacecraft systems, state-awareness, which includes both situational-awareness and self-awareness, is critical for managing the unprecedented amount of uncertainty in knowledge of the state of the systems and the environments to be explored in future missions. This uncertainty introduces significant risk, challenging our ability to validate our systems’ behaviors effectively and decreasing the likelihood that our systems will exhibit correct behaviors at execution time. Endowing our systems with the ability to assess explicitly their state, the state of the environments they operate in, and the associated uncertainties will enable them to make more appropriate and prudent decisions, resulting in greater system resilience. Technologies for state-awareness range from traditional state filters (e.g., Kalman filters) for nominal state estimation and traditional fault protection software (e.g., auto-coded state machines) for off-nominal state diagnosis to more sophisticated model-based estimation and diagnosis capabilities that leverage advances in the field of model-based reasoning.
The spacecraft that support these challenging future missions will need to be capable of reasoning about their own state and the state of their environment in order to predict and avoid hazardous conditions, recover from internal failures, and meet critical science objectives in the presence of substantial uncertainties. System health management (SHM) is the crosscutting discipline of designing a system to preserve assets and continue to function even in the presence of faults. As science goals become more ambitious and our systems are sent to increasingly challenging environments, the simple “safing” response of the past is no longer a viable option. Critical events such as orbit insertion and landing, are extreme examples of the need for more sophisticated responses to faults, since it is impossible to stop the activity and wait for ground operators to diagnose the problem and then to transmit recovery commands; in the time it takes to close the loop with the ground, the opportunity to accomplish the event will have been lost already. Instead, systems must degrade gracefully in harsh environments such as Venus, and they must have the ability to fly through failures in order to complete critical events. Researchers are investigating model-based techniques that will provide a spacecraft with sufficient information to understand its own state and the state of the environment so that it can reason about its goals and work around anomalies when they occur.